Probing and Preparing Novel States of Quantum
Degenerate Rubidium Atoms in Optical Lattices
by
Hirokazu Miyake
Submitted to the Department of Physicsin partial fulfillment of the requirements for the degree of
Doctor of Philosophy
at the
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
June 2013
© Massachusetts Institute of Technology 2013. All rights reserved.
Author . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Department of Physics
May 20, 2013
Certified by. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Wolfgang Ketterle
John D. MacArthur Professor of PhysicsThesis Supervisor
Certified by. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .David E. Pritchard
Cecil and Ida Green Professor of PhysicsThesis Supervisor
Accepted by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .John W. Belcher
Class of ’22 Professor of PhysicsAssociate Department Head for Education
2
Probing and Preparing Novel States of Quantum Degenerate
Rubidium Atoms in Optical Lattices
by
Hirokazu Miyake
Submitted to the Department of Physicson May 20, 2013, in partial fulfillment of the
requirements for the degree ofDoctor of Philosophy
Abstract
Ultracold atoms in optical lattices are promising systems to realize and studynovel quantum mechanical phases of matter with the control and precision offered byatomic physics. Towards this goal, as important as engineering new states of matteris the need to develop new techniques to probe these systems.
I first describe our work on realizing Bragg scattering of infrared light from ul-tracold atoms in optical lattices. This is a detection technique which probes thespatial ordering of a crystalline system, and has led to our observation of Heisenberg-limited wavefunction dynamics. Furthermore, we have observed the superfluid toMott insulator transition through the matter wave Talbot effect. This technique willbe particularly powerful for studying antiferromagnetic phases of matter due to itssensitivity to the crystalline composition.
The second major component of this thesis describes a new scheme to realize theHarper Hamiltonian. The Harper Hamiltonian is a model system which effectivelydescribes electrons in a solid immersed in a very high magnetic field. The effectivemagnetic field manifests itself as a position-dependent phase in the motion of theconstituent particles, which can be related to gauge fields and has strong connectionsto topological properties of materials. We describe how we can engineer the HarperHamiltonian in a two-dimensional optical lattice with neutral atoms by creating alinear potential tilt and inducing Raman transitions between localized states. Insitu measurements provide evidence that we have successfully created the HarperHamiltonian, but further evidence is needed to confirm the creation of the groundstate of this Hamiltonian.
Thesis Supervisor: Wolfgang KetterleTitle: John D. MacArthur Professor of Physics
Thesis Supervisor: David E. PritchardTitle: Cecil and Ida Green Professor of Physics
3
4
Acknowledgments
My time at MIT has been tremendously valuable for me, both professionally as
well as personally. I reckon there are very few institutions in the world where one
is surrounded by so many people with brilliant minds, unwavering passion for what
they do, and the kindest of hearts.
First of all, I thank my Ph.D. supervisors Professor Wolfgang Ketterle and Pro-
fessor Dave Pritchard for their support throughout my time at MIT. I am amazed by
and grateful for the degree of trust Wolfgang put in me from day one to be successful
in lab. The freedom he gave me to thrive in the laboratory has been a very valuable
experience for me. His physical intuition and ability to make connections between
seemingly disparate facts is something I have strived to achieve. Dave has supported
me both as research supervisor in the early years of my Ph.D. as well as an academic
advisor throughout. His deep insight into almost any issue that could be of interest
to the human mind is quite amazing.
Laboratory work can be exhilarating, but if I am completely honest, I have to ad-
mit that there are no shortages of disappointments in running experiments, especially
when one is trying to push back the frontiers of human knowledge. I would not have
been able to get through to the end of my Ph.D. without the help I recieved from my
collaborators in the BEC4 lab. They have been a constant source of encouragement
and support through the ups and downs that is laboratory work. I have also had an
enormous amount of fun with them, day and night, rain or shine. They are a major
reason why my time at MIT has been a fantastic experience.
David Weld was the postdoc when I joined the group, and despite my repeated
questions about basic things, whether it be about atoms, vacuum, or lasers, he pa-
tiently and clearly explained all the questions I would throw at him. His ability to
get things done was very impressive. Without any complaining or whining, he would
get down on his hands and knees to do whatever it takes to make progress in our ex-
periment. His perseverance and calmness is something which I have tried to emulate.
I am very fortunate to have had him as an immediate superior during the beginning
5
of my Ph.D. Also he was one of very few people I could talk to about my interest in
aviation, and I hope to one day fly on a A380 and brag to him about it.
Patrick Medley was the senior graduate student when I joined the group. I can
say that I have never met a person with such a fast mind. Despite differences in
information processing speed between the two of us, we were able to form a very good
friendship, and shared many interests. I have to say that playing old school video
games with him was one of the highlights of my time at MIT. I do not think I had as
much fun with somebody since way, way back. And then all the various episodes we
watched were also good times. In particular, I am proud to have witnessed with him
a historic event as it happened, which was the end of ‘Endless Eight.’ Late nights in
lab were made bearable by all that Patrick had to offer.
Georgios Siviloglou was the second postdoc that I interacted with, and he joined
about midway through my Ph.D. We immediately became good friends. His under-
standing of and appreciation for the big picture, both in research and in life provided
a lot of food for thought. In particular I was very impressed by his understanding
and mastery of ‘social engineering’ and I am sure his influence shows through in the
writing of this thesis. I will remember our stroll through the Toy District in Los
Angeles at dusk as an interesting adventure. I will miss the coffee breaks at Area
Four. I am looking forward to reading the Illiad upon graduation and hopefully I will
have a chance to visit Mount Athos someday.
David Hucul was one year above me when I entered MIT. His command of elec-
tronics was nothing short of amazing to me, and his optimism and positive outlook
helped to maintain good morale in the lab. I am also grateful to him for inviting me
and my co-workers to my first ever commercial shoot, which was a fun experience.
It taught me that I should work on my acting before I give it another try. Claire
Leboutiller joined the BEC4 lab for a few months and worked on putting together
Helmholtz coils for our science chamber. It was a pleasure working with her, and I
admire her ability to get the job done with minimal supervision. Yichao Yu joined
the BEC4 lab only when he was a sophomore, but his command of physics was ev-
ident from early on. His support on the Bragg results presented in this thesis was
6
invaluable, and I am sure he will go on to do great things.
After I leave, Colin Kennedy will be the senior student on the BEC4 experiment.
He is very lab smart, and only in a short time he has taken initiative numerous
times and pushed forward many upgrades to our experiment for the better. With his
willingness to take action, combined with his confidence, I am sure he will take the
lab to new heights of success. Cody Burton is the youngest member in the BEC4
lab. Although he has been around for less than one year, it is clear that his calm and
focus will be a valuable asset to the future operation of BEC4. I am looking forward
to hearing great news from this lab as Colin and Cody ‘take it to the next level.’
No acknowledgement will be complete without mentioning the other people in the
Center for Ultracold Atoms. The collective knowledge they hold and the amount of
supplies they provide in cases of emergencies have been an absolutely crucial part
of our work. I am thankful for Christian Sanner for providing excellent technical
help during the oddest hours, such as fixing puzzling problems with our magneic
trap. Aviv Keshet is the unsung hero of the hallway, perhaps even the entire field
of ultracold atoms. His Cicero experimental control program has provided us with a
versatile and powerful platform to control our experiments and many other experi-
ments throughout the world. Ed Su and I entered graduate school at the same time.
His knowledge of ultracold atomic physics intimidated me at first, but he was also
a great guy and his excitement for physics was contageous. Ralf Gommers had an
excellent command of technical matters, and his soccer skills were also admirable.
Jonathon Gillen deserves my thanks for reviewing our Bragg scattering paper before
we sent it off for publication. I also appreciate him sharing his experience in the
patent law field. Wujie Huang is a wonderfully nice person, and his insights about
Raman-assisted tunneling helped to refine my understanding and helped make this
thesis better. I want to thank Professor Martin Zwierlein for agreeing to serve on
my thesis committee, as well as for his fighting spirit which kept the morale of our
intramural soccer team high. Peyman Ahmadi deserves a profound thank you for
many helpful technical discussions. His knowledge was truly valuable in solving our
own problems, and he was always gracious with his time to help us. He was truly one
7
of the kindest person in the hallway, in addition to being the greatest goal keeper the
CUA has ever known. Cheng-Hsun Wu deserves my thanks for hosting me during the
MIT Physics Department open house when I was choosing which graduate school to
attend. Ivana Dimitrova, Niklas Jepsen and Jesse Amato-Grill also deserve a thank
you for allowing us to borrow many of their equipments, and I am glad we shared an
office for a brief time. Their joining the group definitely made the hallway a much
more lively place. I appreciated the discussions I had with Haruka Tanji-Suzuki about
our progress in graduate school and how every single time we talked, we could not
belive so much time had elapsed since we talked last time. I thank all the people
I met at conferences and summer schools, in particular at Les Houches in 2009. I
will always remember the mountain hiking in the Alps, the breathtaking views, and
that one pear in the lunch sack I ate after hiking for hours, which was the most deli-
cious pear I ever had, or may ever have. I also want to thank everybody who played
in the MIT intramural soccer team ‘Balldrivers,’ in particular Sebastian Hofferberth
and Andrew Grier for helping me out as captain, and everybody who played in the
CUA softball team against the JILA team at various DAMOP conferences. Joanna
Keseberg deserves a profound thank you for all her help in dealing with administra-
tive matters in a timely and professional way. She is what makes the CUA run as
smoothly as it does.
There are also people outside of the lab which deserve special mention. Riccardo
Abbate, Francesco D’Eramo and Thibault Peyronel have been good friends since the
beginning of my time at MIT and we had fun times studying for our qualifying exams.
I also want to thank Emily Florine, Michelle Sukup Jackson, Monica Sun and James
Sun for fun times outside of physics during my first year here when I did not know
too many people. Ariel Sommer and Matt Schram have been good roommates at
Ashdown since the second year. I thank Steph Teo and Been Kim for good times in
the later part of my studies. I want to thank Stephanie Houston for being a great
pottery teacher. I also want to thank the students from all the seminar XL classes
I have taught, as well as the staff at the Office of Minority Education for giving me
the opportunity to teach undergraduates all throughout my time at MIT. I want
8
to thank the wonderful friends I met at Cornell University who have continued to
be good friends long after I have left Cornell, including Carol-Jean Wu, Linda Wu,
Mark Wu, Jah Chaisangmongkon, Ava Wan, Joyce Lin, Thiti Taychatanapat, Will
Cukierski, Jim Battista, Joe Calamia, Saul Blanco Rodrıguez and Nabil Iqbal.
I would also like to acknowledge the physicists I met during my undergraduate
studies at Cornell University that convinced me to go to graduate school in physics. I
am grateful to Professor Anders Ryd for taking me on as an undergraduate researcher
during my sophomore year despite the fact that I knew nothing about research. He
very patiently guided me through the research process of high energy experiments, in
particular analysis of data from the CLEO-c detector located at Cornell University
and helped me publish my first scientific paper. Needless to say, this gave me a lot
of satisfaction and confidence in continuing in my physics research career. I had the
good fortune of interacting with many knowledgeable and great people that were a
part of the CLEO collaboration, and in particular I want to thank Werner Sun and
Peter Onyisi for patiently guiding me through my data analysis. I am also grateful
to Professor J. C. Seamus Davis for allowing me to join his research group. His
very engaging lectures on introductory quantum mechanics got me interested in his
work on quantum matter. The opportunity to work on measuring short-length-scale
gravity with superconducting quantum interference devices was definitely a fantastic
experience which taught me the excitement of doing research to probe the cutting-
edge of what is possible. It was a joy interacting with the various members of the
Davis group, and in particular I want to acknowledge Benjamin Hunt, Ethan Pratt,
and Minoru Yamashita for introducing me to working with cryogenics and teaching
me how to build things right, all the way from the design stage to the smallest details
such as how to solder a wire onto a board in the best way possible. I learned a
tremendous amount from all of these people.
I have also had the good fortune of being able to explore different aspects of science
here at MIT. I was fortunate to be able to participate in Congressional Visits Day
organized by the MIT Science Policy Initiative, which allowed me to experience first-
hand how policy is made in the U.S. and how scientists can help the government make
9
better decisions. I was also able to take courses on ‘Managing Nuclear Technology’ by
Professor Richard Lester, ‘Law and Cutting-Edge Technologies’ by John Akula, and
‘Nuclear Forces and Missle Defenses’ by Professor Ted Postol. I learned a lot from
all of these professors and they helped expand my understanding of how science and
technology can affect society and vice versa. In particular, I want to acknowledge the
advices Professor Postol gave me as I was considering my next position after graduate
school. His lectures inspired me to make use of the technical understanding that I
had developed through my studies of physics towards the betterment of society.
I want to thank my family, Itsuo, Kazuko, Maria, Lisa, and Julia who have sup-
ported me throughout my life and allowed me to pursue the path I wanted to take. I
could not have asked for better parents and sisters.
Finally, I want to thank Nicole Lim and the entire Lim and Sy family for embracing
me as part of their family. Our adventure is only beginning.
10
Contents
1 Introduction 17
1.1 Quantum Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.2 Motivations and Outline . . . . . . . . . . . . . . . . . . . . . . . . . 19
2 Ultracold Atoms in Optical Lattices 21
2.1 Optical Lattice Potential . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2 Atoms in a Periodic Potential . . . . . . . . . . . . . . . . . . . . . . 23
2.2.1 Bloch’s Theorem and Bloch Functions . . . . . . . . . . . . . 24
2.2.2 Wannier Functions . . . . . . . . . . . . . . . . . . . . . . . . 26
2.2.3 Wannier-Stark States . . . . . . . . . . . . . . . . . . . . . . . 27
2.3 Bose-Hubbard Hamiltonian and the Mott Insulator . . . . . . . . . . 29
2.3.1 Mott Insulator Phase Diagram . . . . . . . . . . . . . . . . . . 33
2.3.2 Non-Zero Temperature Effects . . . . . . . . . . . . . . . . . . 35
2.4 Heisenberg Hamiltonian and Quantum Magnetism . . . . . . . . . . . 36
2.4.1 Anisotropic Heisenberg Hamiltonian . . . . . . . . . . . . . . 38
2.4.2 Quantum Magnetism and Bragg Scattering . . . . . . . . . . . 40
2.5 Harper Hamiltonian and Gauge Fields . . . . . . . . . . . . . . . . . 41
2.5.1 Electromagnetism, Gauge Fields, and Wavefunctions . . . . . 43
2.5.2 Flux Density per Plaquette α . . . . . . . . . . . . . . . . . . 45
2.5.3 Band Structure for α = 1/2 . . . . . . . . . . . . . . . . . . . 46
3 Bragg Scattering as a Probe of Ordering and Quantum Dynamics 49
3.1 Experimental Setup for Bragg Scattering . . . . . . . . . . . . . . . . 50
11
3.1.1 Characterization of Bragg Scattering . . . . . . . . . . . . . . 52
3.2 Debye-Waller Factor and Bragg Scattering . . . . . . . . . . . . . . . 56
3.3 Heisenberg-Limited Wavefunction Dynamics . . . . . . . . . . . . . . 60
3.4 Superfluid Revivals of Bragg Scattering . . . . . . . . . . . . . . . . . 61
3.5 Superfluid to Mott Insulator Through Bragg Scattering . . . . . . . . 62
3.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4 Raman-Assisted Tunneling and the Harper Hamiltonian 67
4.1 Raman-Assisted Tunneling in the Perturbative Limit . . . . . . . . . 70
4.1.1 Energy Hierarchy to Realize the Harper Hamiltonian . . . . . 73
4.2 Experimental Geometries to Realize Specific α . . . . . . . . . . . . . 74
4.2.1 Experimental Geometry for α = 1/2 . . . . . . . . . . . . . . 75
4.3 Raman-Assisted Tunneling in the Rotating Frame . . . . . . . . . . . 75
4.3.1 Perturbative and Exact Raman-Assisted Tunneling . . . . . . 83
4.4 Amplitude and Phase Modulation Tunneling . . . . . . . . . . . . . . 83
4.5 Experimental Studies of Amplitude Modulation (AM) Tunneling . . . 86
4.5.1 Experimental Sequence for AM Tunneling . . . . . . . . . . . 86
4.5.2 In Situ Measurements of AM Tunneling . . . . . . . . . . . . 87
4.5.3 Time-of-Flight Measurements of AM Tunneling . . . . . . . . 90
4.6 Experimental Observation of Raman-Assisted Tunneling . . . . . . . 91
4.6.1 Experimental Sequence for Raman-Assisted Tunneling . . . . 91
4.6.2 In Situ Measurements of Raman-Assisted Tunneling . . . . . . 92
4.6.3 Time-of-Flight Measurements of Raman-Assisted Tunneling . 97
4.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
5 Future Prospects 99
5.1 Prospects for Spin Hamiltonians . . . . . . . . . . . . . . . . . . . . . 99
5.2 Prospects for Topological Phases of Matter . . . . . . . . . . . . . . . 100
5.3 Prospects for Quantum Simulation . . . . . . . . . . . . . . . . . . . 101
A Optical Lattice Calibration Using Kapitza-Dirac Scattering 103
12
B Code to Construct Wannier Functions 109
C Bragg Scattering as a Probe of Atomic Wave Functions and Quan-
tum Phase Transitions in Optical Lattices 113
D Spin Gradient Demagnetization Cooling of Ultracold Atoms 119
E Thermometry and refrigeration in a two-component Mott insulator
of ultracold atoms 125
F Spin Gradient Thermometry for Ultracold Atoms in Optical Lat-
tices 131
Bibliography 137
13
14
List of Figures
2-1 Band structure for an atom in an optical lattice . . . . . . . . . . . . 26
2-2 Wannier-Stark states in a tilted lattice . . . . . . . . . . . . . . . . . 28
2-3 Dependence of Bose-Hubbard Hamiltonian parameters on lattice depth 32
2-4 Phase diagram of the superfluid to Mott insulator transition. . . . . . 34
2-5 Average occupation number as a function of chemical potential . . . . 37
2-6 Phase diagrams of the Heisenberg Hamiltonian . . . . . . . . . . . . . 41
2-7 Hofstadter butterfly . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2-8 Setup for the Aharonov-Bohm effect . . . . . . . . . . . . . . . . . . . 44
2-9 Phase accumulated upon tunneling and enclosed flux . . . . . . . . . 46
2-10 Schematic of a two-dimensional lattice with phases . . . . . . . . . . 47
2-11 Band structure of the α = 1/2 Harper Hamiltonian . . . . . . . . . . 48
3-1 Bragg scattering schematic setup . . . . . . . . . . . . . . . . . . . . 51
3-2 Bragg reflection as a function of the guiding mirror screw angle . . . 53
3-3 Bragg scattering as a function of Bragg probe power . . . . . . . . . . 55
3-4 Bragg scattering as a function of Bragg probe time . . . . . . . . . . 55
3-5 Bragg reflection as a function of razor position . . . . . . . . . . . . . 57
3-6 Bragg scattering as a probe of the spatial wavefunction width . . . . 58
3-7 Bragg scattering as a probe of the momentum wavefunction width . . 60
3-8 Revivals of Bragg scattered light . . . . . . . . . . . . . . . . . . . . . 63
3-9 Bragg scattering revival as a probe of superfluid coherence . . . . . . 64
4-1 Schematic of a tilted lattice and Raman beams. . . . . . . . . . . . . 69
4-2 Overlap integrals for perturbative Raman coupling . . . . . . . . . . . 72
15
4-3 Tilt and phase accumulated for general Raman geometry . . . . . . . 75
4-4 Experimental geometry for α = 1/2 . . . . . . . . . . . . . . . . . . . 76
4-5 Tilt and Wannier-Stark states . . . . . . . . . . . . . . . . . . . . . . 77
4-6 Overlap integrals to calculate Raman-assisted tunneling . . . . . . . . 79
4-7 Perturbative and exact treatment of Raman-assisted tunneling . . . . 84
4-8 In situ amplitude modulation tunneling measurements . . . . . . . . 88
4-9 Evolution of superfluid peaks under lattice amplitude modulation . . 90
4-10 Raman-assisted tunneling setup and spectrum . . . . . . . . . . . . . 92
4-11 Raman-assisted tunneling leads to linear expansion in time . . . . . . 93
4-12 Raman-assisted tunneling for high Raman beam intensities . . . . . . 94
4-13 Raman-assisted tunneling as a function of the lattice depth . . . . . . 95
4-14 Higher-order and Next-nearest-neighbor tunneling with Raman beams 97
B-1 Wannier function for lattice spacing 532 nm and lattice depth 3ER . . 112
16
Chapter 1
Introduction
The realization of Bose-Einstein condensates (BECs) in ultracold atomic sys-
tems [7, 23] opened a new direction in the study of macroscopic quantum phenomena.
These include the observation of matter wave interference [8], spinor dynamics [99],
and vortex lattices [1], significantly deepening our understanding of quantum mat-
ter. It was clear from early on in the experimental exploration of BECs that there
were deep connections between these atomic systems and those studied in condensed
matter systems [62]. These include superfluidity as observed in liquid helium and
superconductivity in various materials. At a fundamental level, many systems that
were only accessible in condensed matter systems were now also open for investigation
in atomic systems, but with different control parameters.
Some of the relevant properties and techniques that are unique to atomic systems
include the existence of Feshbach resonances which allow experimentalists to tune the
interaction energy between atoms [55], optical dipole traps to trap and manipulate
atoms [95], and the ability to control the dimensionality of the system between three
dimensions, two dimensions, and one dimensions [41]. Such controls allowed the
exploration of quantum matter in systems which were complementary to what had
been studied in condensed matter. The creation of quantum degenerate Fermi gases
further diversified the type of systems realizable [27], allowing atomic physicists to
study both bosonic as well as fermionic particles.
One major addition to the controllability and flexibility already discussed is the
17
ability to genereate an optical lattice potential to create a crystalline structure analoglous
to condensed matter systems [70, 13]. This has allowed the field to move in the di-
rection of creating systems directly analogous to novel solid state materials such as
high-temperature superconductors and magnetic phases of matter.
1.1 Quantum Simulation
Quantum simulation was first articulated by Richard Feynman [32], and the idea is
that you can create a quantum system with exactly the same parameters as another
one that is more controllable, and study what would have happened to the other
system. Groundbreaking work on the superfluid to Mott insulator transition with a
three-dimensional optical lattice [42] was an early success of how ultracold atoms in
optical lattices can realize interesting many body physics as those found in condensed
matter, where interactions between constituent particles in a lattice can be important.
Another example is the Fermi-Hubbard Hamiltonian [53], which in condensed
matter describes electrons in a crystal which can undergo a transition from a con-
ducting state to an insulating state and is of particular interest because of the belief of
its strong connection to high-temperature superconductors in condensed matter [69].
The exact same Hamiltonian has been realized in ultracold atomic physics, where a
fermionic atom resides in a three-dimensional crystal structure created by interfering
light beams [88, 61]. Thus the idea is that by studying the Fermi-Hubbard Hamilto-
nian in atomic systems, physicists can gain a better understanding of the mechanisms
behind high-temperature superconductors in condensed matter. Note the fact that
the atoms must be fermionic in this case to more accurately simulate a condensed
matter system made of electrons.
Fermions have a direct analogy to electrons in real condensed matter systems,
whereas fundamentally bosonic particles do not exist in condensed matter. However,
many interesting quasiparticles in condensed matter are bosonic in nature, including
Cooper pairs in superconductors [20], excitons in semiconductors [119], and magnons
in magnetic materials [14]. Thus although the constitutive particles in condensed
18
matter are not bosons, understanding bosonic particles in ultracold atomic systems
can significantly add to our understanding of quantum matter. From a theoretical
point of view, bosonic systems are typically easier to calculate, so a precise comparison
between theory and experiment can be possible [106]. Also, validating experimental
techniques with bosonic systems where more experimental tools and understanding
are available is invaluable to exploring fermionic systems [11, 90, 77]. Finally, the non-
existence of fundamentally bosonic particles in condensed matter makes ultracold
bosonic systems unique in that systems which were previously only of theoretical
importance can now be studied experimentally [29].
1.2 Motivations and Outline
The rubidium lab at MIT which I belong to and where the work described in
this thesis was done uses ultracold 87Rb atoms [100], which are bosonic, and has
been focusing on studying these atoms in optical lattices which closely mimic crystal
structure in condensed matter. In particular our major goal has been to realize new
magnetic forms of matter in its various guises.
One direction has been to realize and observe phase transitions to ferromagnetic
and antiferromagnetic phases [85]. The fermionic antiferromagnet is expected to have
deep connections with high-temperature superconductors, but the bosonic antiferro-
magnet is also expected to be interesting in its own right [39]. Towards that goal
this lab has developed the technique of spin gradient thermometry and spin gradient
demagnetization cooling, which has allowed us to measure and cool our system down
to 300 pK, putting us in striking range of these magnetic states [115, 116, 76]. I was
involved in the thermometry and cooling experiments in the first half of my Ph.D.
studies, and the published papers are presented in appendices D, E, and F. In this
thesis I describe our work on Bragg scattering applied to ultracold atoms, which will
allow us to study the antiferromagnetic state once it is created.
Another direction has been to realize Hamiltonians which can be described by
gauge fields [22]. These are analoglous to electrons in a magnetic field in condensed
19
matter which gives rise to such phenomena as the quantum hall effect [107, 68, 120].
One important parameter for such systems is the magnetic flux enclosed per area,
where the area is determined by the relevant length scale of the system. In condensed
matter, the necessary magnetic fields to realize magnetic fluxes of order the magnetic
flux quanta where interesting effects are expected to occur are at around a billion
gauss due to the very small length scales involved of order angstrom [52], which is
unachievable in the laboratory. However, with cold atoms it may be possible to
realize such effectively high magnetic fluxes [59, 37]. Much experimental progress has
been made towards realizing and studying artificial gauge fields in ultracold atomic
systems [72, 3, 101], but simple and robust schemes to achieve high magnetic fluxes
have not yet been realized. In this thesis I describe a setup to realize very high
magnetic fields which can give rise to such fascinating features as the Hofstadter
butterfly, which is a fractal band structure.
This thesis is organized as follows. In chapter 2, I describe some of the basics
of optical lattices, as well as the behavior of particles in a sinusoidal potential. The
chapter also discusses the Bose-Hubbard Hamiltonian, Heisenberg Hamiltonian, and
the Harper Hamiltonian. In chapter 3, I describe our work on Bragg scattering to
study the Bose-Hubbard Hamiltonian and atomic dynamics. In chapter 4, I describe
the scheme we have developed to realize artificial magnetic fields as well as experi-
mental results to realize the Harper Hamiltonian. Then chapter 5 offers conclusions
and outlooks.
20
Chapter 2
Ultracold Atoms in Optical
Lattices
A coherent laser beam is a particularly versatile tool to control dilute, ultracold
atomic systems due to their spectral purity as well as the controlability in spatial and
temporal structure. In particular laser beams are standard tools used to trap and
manipulate atoms to create new states of matter.
In this chapter I describe the principles behind optical lattices, as well as some
of the properties of atoms in an optical lattice potential and the different types of
Hamiltonians one can realize, including the Bose-Hubbard Hamiltonian, one of the
simplest ultracold lattice systems which exhibit strong interaction effects, the Heisen-
berg Hamiltonian, which describes magnetic phases of matter including ferromagnets
and antiferromagnets, and the Harper Hamiltonian, which exhibits topological prop-
erties of matter.
2.1 Optical Lattice Potential
Atoms that are sufficiently cold can by trapped at the focus of a Gaussian laser
beam which induces a light shift if red-detuned to an atomic transition, thereby
creating a potential minimum, which corresponds to an intensity maximum for the
cold atoms to sit in [19, 96].
21
For an atom in an oscillating electric field E(t) = Ee cosωt, the perturbing Hamil-
tonian H ′(t) can be written
H ′(t) = −d · E(t). (2.1)
The energy shift due to an oscillating electric field is given by
U =1
2〈H ′(t)〉 (2.2)
= −∑k
E2|Dkg|24~
(− 1
δkg+
1
ωkg + ω
)(2.3)
where Dkg ≡ e〈k|z|g〉, δkg ≡ ω − ωkg, and I = cε0E2/2.
So far we have only considered a single coherent laser beam, but if we have two
such beams, they can interfere and create a standing wave intensity pattern. The
resulting sinusoidal potential which the atoms see is called an optical lattice.
Consider two plane electromagnetic waves E1(r, t) and E2(r, t) that are propagat-
ing in arbitrary directions k1 and k2 respectively, where E1(r, t) = E0ε1ei(k1·r−ωt) and
E2(r, t) = E0ε2ei(k2·r−ωt+φ), where ε1,2 are polarization vectors, E0 is the electric field
magnitude, k1,2 are the wavevectors, and φ is an arbitrary constant phase. The total
electric field is given by Etot(r, t) = E1(r, t) + E2(r, t). Then we have
|Etot(r, t)|2 = E20 |ε1e
ik1·r + ε2ei(k2·r+φ)|2 (2.4)
= E20(2 + ε1 · ε∗2ei(k1·r−k2·r−φ) + ε∗1 · ε2e
−i(k1·r−k2·r−φ)) (2.5)
= 2E20(1 + ε1 · ε2 cos((k1 − k2) · r− φ)), (2.6)
where we assumed the polarization vectors to be real. Now consider the case when the
polarization vectors are parallel so that ε1 ·ε2 = 1. Then, from cos2 θ = (1+cos 2θ)/2,
22
we get
|Etot(r, t)|2 = 4E20 cos2
((k1 − k2) · r− φ
2
)(2.7)
= 4E20 cos2
(keff · r−
φ
2
)(2.8)
where we define keff ≡ (k1 − k2)/2. If we define θ such that k1 · k2 = cos θ and using
k = 2π/λ, we get
λeff =1
sin(θ/2)λ, (2.9)
where keff ≡ 2π/λeff . Thus, by intersecting laser beams at various angles, one can
create an optical lattice with lattice spacings much greater than the wavelength of the
laser beam itself. Also, note that the direction of the optical lattice will be modulated
in the direction k1 − k2.
In the work described in this thesis θ = π, so λeff = λ and we have
|Etot(r, t)|2 = 4E20 cos2
(k · r− φ
2
). (2.10)
Thus this periodic modulation in the intensity pattern creates an optical lattice,
which the atoms reside in. The next section describes the properties of atoms in such
a sinusoidal potential. Typically the lattice laser beams are significantly bigger than
the optical dipole trap beams, so the spatial curvature due to the Gaussian envelope
of the lattice beams can be neglected.
2.2 Atoms in a Periodic Potential
In 1D, a single atom in a standing wave is described by the Hamiltonian
H =p2
2m+ V0 sin2(klx), (2.11)
23
where m is the mass of the particle, and kl = 2π/λ for an optical lattice with laser
wavelength λ.
2.2.1 Bloch’s Theorem and Bloch Functions
Bloch’s theorem [9] tells us that if the Hamiltonian is invariant under translation
by one lattice period a, the Hamiltonian commutes with the translation operator T ,
where
T = eipa/~ and Tφq(x) = φq(x+ a). (2.12)
Since T is unitary, it has eigenvalues such that
Tφq(x) = eiqaφq(x), (2.13)
where q ∈ [−π/a, π/a]. Since
φq(x+ a) = eiqaφq(x), (2.14)
we can write
φq(x) = eiqxuq(x), (2.15)
where uq(x) is a periodic function in x with period a i.e. uq(x) = uq(x+ a). Because
the Hamiltonian and the translation operator commute, we can find simultaneous
eigenstates of H and T such that
Hφ(n)q (x) = E(n)
q φ(n)q (x) and (2.16)
Tφ(n)q (x) = eiqaφ(n)
q (x). (2.17)
24
φnq (x) are called Bloch eigenstates, and unq (x) are eigenstates of the Hamiltonian
Hq =(p+ q)2
2m+ V0 sin2(klx). (2.18)
This follows from using the fact that
e−iqxpeiqx = p+ q. (2.19)
We would now like to determine the Bloch functions, which are normalized so that
2π
a
∫ a
0
|φ(n)q (x)|2dx = 1. (2.20)
The analytical solutions are of the form of Mathieu functions. However, numerically
it is nicer to expand the Bloch functions in a Fourier expansion, where
unq (x) =1√2π
∞∑j=−∞
c(n,q)j e−iklxj. (2.21)
This allows us to reduce the problem to a linear eigenvalue equation in the complex
coefficients cj. To do this, take Schrodinger’s equation and use
1√2π
∫ ∞−∞
dxe−ikx =√
2πδ(k) and δ(ax) =1
|a|δ(x). (2.22)
By doing so we get
l∑j′=−l
Hj,j′c(n,q)j′ = E(n)
q c(n,q)j , (2.23)
where
Hj,j′ =
(2j + q/kl)
2ER + V0/2, if j = j′;
−V0/4, if |j − j′| = 1;
0 if |j − j′| > 1.
(2.24)
25
−1 −0.5 0 0.5 10
2
4
6
8
10
12
14
16
qa/π
E/E
R
Eigenvalues of Bloch States
n=0n=1n=2n=3
−1 −0.5 0 0.5 10
2
4
6
8
10
12
14
16
18
20
qa/π
E/E
R
Eigenvalues of Bloch States
n=0n=1n=2n=3
−1 −0.5 0 0.5 12
4
6
8
10
12
14
16
18
20
22
qa/π
E/E
R
Eigenvalues of Bloch States
n=0n=1n=2n=3
Figure 2-1: Band structure for an atom in an optical lattice. From left to right arethe eigenvalues for an atom in a lattice for lattice depths V0 = 0ER, V0 = 5ER, andV0 = 10ER as a function of the quasimomentum q.
This problem can be diagonalized numerically using j ∈ −l, . . . ,−l and for the
lowest few bands, good results are obtained for relatively small l, i.e. l ∼ 10. The
numerically determined eigenvalues for the first few energy bands are given in Fig. 2-1.
2.2.2 Wannier Functions
It is often very convenient to write the Bloch functions in terms of Wannier func-
tions, which also form a complete set of orthogonal basis states. The Wannier func-
tions in 1D are given by
wn(x− xi) =
√a
2π
∫ π/a
−π/adqu(n)
q (x)e−iqxi , (2.25)
where xi are the minima of the potential. Each set of Wannier functions for a given
n can be used to express the Bloch functions in that band as
u(n)q (x) =
√a
2π
∑xi
wn(x− xi)eiqxi . (2.26)
This can be checked by plugging Eq. 2.26 into Eq. 2.25. Wannier functions have
the advantage of being localized on particular sites, which makes them useful for de-
scribing local interactions between particles. The Wannier functions are not uniquely
defined by the integral over Bloch functions since each wavefunction φ(n)q (x) is arbi-
trary up to a complex phase. However, it has been shown [65] that there exists for
26
each band only one real Wannier function wn(x) that is
either symmetric or anti-symmetric about either x = 0 or x = a/2 and
falls off exponentially i.e. |wn(x)| ∼ exp(−hnx) for some hn > 0 as x→∞.
These Wannier functions are known as maximally localized Wannier functions, and
are the ones frequently used. If u(n)q (x) is expanded as in Eq. 2.21, the maximally
localized Wannier functions can be produced if all c(n,q)j are chosen to be purely
real for the even bands i.e. n = 0, 2, 4, . . . , and imaginary for the odd bands i.e.
n = 1, 3, 5, . . . , and are smoothly varying as a function of q. Wannier functions
for deeply bound bands are very close to the harmonic oscillator wavefunctions, and
for many analytical estimates of on-site properties, the Wannier functions may be
replaced by harmonic oscillator wavefunctions if the lattice is sufficiently deep. The
major difference between the two is that the Wannier functions are exponentially
localized, i.e. |wn(x)| ∼ exp(−hnx), whereas the harmonic oscillator wavefunctions
decay more rapidly in the tails as exp−x2/(2a0)2.
2.2.3 Wannier-Stark States
Now consider a situtation where in addition to the periodic lattice potential we
have a linear lilt potential due to an effective constant time-independent force F such
that we obtain the Wannier-Stark Hamiltonian [45, 40]
H =p2
2m+ V0 sin2(klx) + Fx. (2.27)
Note that the linear tilt destroys the translation symmetry of the tiltless Hamiltonian,
and so the Wannier wavefunctions will not be the proper description of a localized
state. Being able to describe the localized wavefunctions resulting from this Hamil-
tonian, called Wannier-Stark states, is important in understanding our scheme for
engineering the Harper Hamiltonian in an ultracold atomic system which is further
described in chapter 4.
27
−4 −3 −2 −1 0 1 2 3 4−1000
−500
0
500
1000
1500
2000
2500
Position (λ/2)
Wav
efun
ctio
n an
d P
oten
tial (
arb.
)
Figure 2-2: Wannier-Stark states in a tilted lattice.
In a sufficiently deep lattice such that the tight-binding approximation can be
used, and a sufficiently steep tilt such that the bare tunneling energy J (this parameter
is explained in the following section) is smaller than the energy offset between adjacent
sites Fa, the Wannier-Stark Hamiltonian can be diagonalized in the basis of the
Wannier-Stark states |Ψα,l〉 given by
|Ψα,l〉 =∑m
Jm−l
(2J
Fa
)|m〉, (2.28)
where |m〉 are Wannier functions centered at site m, and Jm−l(x) are Bessel functions.
Thus when one is interested in the properties of a lattice system in the presence of a
linear tilt potential, the Wannier-Stark states should be used instead of the Wannier
states. Wannier-Stark states are depicted in Fig. 2-2.
28
2.3 Bose-Hubbard Hamiltonian and the Mott In-
sulator
Although band theory for non-interacting particles has been hugely successful in
explaining the properties of a wide range of metals [9], there are a few shortcomings.
In particular, there are a class of materials, including NiO, which band theory would
predict should be conductors, but are in reality insulators [24]. The Hubbard model
proposed by John Hubbard has been shown to explain quite well why this material
is insulating [53, 63]. In particular, this model Hamiltonian captures the interactions
between the electrons effectively in the tight-binding approximation and is a useful
model to begin to understand the effects of interactions between constitutent particles
in a lattice system.
The Bose-Hubbard Hamiltonian also describes a quantum phase transition be-
tween a superfluid state and an insulating state, called the Mott insulator [33, 58].
This was first observed in an ultracold atomic system in 2002 [42] and has been stud-
ied extensively since then. In chapter 3 of this thesis I describe our work on studying
this quantum phase transition using the technique of Bragg scattering. The Mott
insulating state is also the starting point in realizing spin gradient thermometry and
demagnetization cooling, where the published papers [76, 116, 115] are included in
this thesis as appendices D, E, and F.
Now in an ultracold atomic system, for bosonic atoms in a lattice and an external
trapping potential, the Hamiltonian operator is given by [58]
H =
∫d3xψ†(x)
(− ~2
2m∇2 + V0(x) + VT (x)
)ψ(x)
+1
2
4πa~2
m
∫d3xψ†(x)ψ†(x)ψ(x)ψ(x), (2.29)
where ψ(x) is a boson field operator for atoms in a given internal atomic state, V0(x)
is the optical lattice potential, and VT (x) is the additional slowly varying external
trapping potential, such as a magnetic or optical dipole trap. In the simplest case,
29
the optical lattice potential has the form V0(x) =∑3
j=1 Vj0 sin2(kxj) with wavevector
k = 2π/λ where λ is the wavelength of the lattice beam. The interaction potential
between the atoms is approximated by a short-range pseudopotential where a is the
s-wave scattering length and m is the mass of the atom. If we assume the energies of
the system are low enough that the atoms remain in the lowest band, we can expand
the field operators in the Wannier basis and keep only the lowest vibration states and
write ψ(x) =∑
i aiw(x− xi). This then allows us to simplify Eq. 2.29 to obtain the
Bose-Hubbard Hamiltonian
H = −J∑〈i,j〉
a†iaj +1
2U∑i
ni(ni − 1) +∑i
εini, (2.30)
where 〈i, j〉 denotes a sum over the nearest neighbor sites, the operators ni = a†iai
count the number of bosonic atoms at site i. The annihilation and creation operators,
ai and a†i respectively, obey the canonical commutation relations [ai, a†j] = δij. The
parameter U = 4πa~2∫d3x |w(x)|4 /m corresponds to the strength of the on-site
repulsion of two atoms on the lattice site i, a is the s-wave scattering length of the
atom, J =∫d3xw∗(x− xi)[− ~2
2m∇2 + V0(x)]w(x− xj) corresponds to the hopping
matrix element between adjacent sites i and j, and εi =∫d3xVT (x)|w(x− xi)|2 ≈
VT (xi) corresponds to an energy offset of each lattice site.
One can use a harmonic approximation about the minima of the potential wells
to determine these parameters. In particular, for 1D,
H ′ =p2
2m+ V0 sin2(kx) ≈ p2
2m+ V0k
2x2 (2.31)
=p2
2m+
1
2mω2x2 (2.32)
⇒ ω2 =2V0k
2
m=
4ERV0
~2⇒ ω =
√4ERV0
~2, (2.33)
where ER = ~2k2/(2m) is the recoil energy. Furthermore, the ground state eigenfunc-
30
tion of this Hamiltonian can be written
w(x) =1
(πσ2)1/4e−
x2
2σ2 , (2.34)
where
σ =
√~mω
=~
m1/2(4ERV0)1/4. (2.35)
The generalization to 3D is given by
w(x) =1
π3/4(σ1σ2σ3)1/2e−(x2
2σ21+ y2
2σ22+ z2
2σ23
). (2.36)
Using the expression for U and the above approximate Wannier function, we may
write
U =4πa~2
m
∫d3x |w(x)|4 (2.37)
≈√
8
πka (ERV1V2V3)1/4 . (2.38)
In all experiments described in this thesis, we work with lattice lasers of wavelength
1064 nm and 87Rb atoms which weigh 1.443× 10−25 kg with s-wave scattering length
of 100a0 (in reality, the scattering length depends on the hyperfine states that are
undergoing the scattering, but for 87Rb all relevant scattering lengths are identi-
cal to within about 1% and so differences are typically negligible) [116]. For these
parameters, Fig. 2-3 shows the interaction energy U , tunneling energy J , and the
superexchange energy J2/U as a function of lattice depth in units of the recoil energy
ER. These three terms set the energy scale of the system.
The on-site interaction energy term can be most naturally be understood as an
energy between any two atoms on a single lattice site. In particular, if a lattice site
has n atoms in it, then there are n choose 2 pairs that one can form, and each of
31
0 2 4 6 8 10 12 14 16 18 2010
−2
10−1
100
101
102
103
104
Lattice Depth (ER)
Ene
rgy
(Hz)
JU
J2/U
Figure 2-3: Dependence of Bose-Hubbard Hamiltonian parameters on lattice depth.Calculations are done for a 3D optical lattice of equal lattice depth in all three di-rections created by a laser of wavelength 1064 nm and 87Rb atoms which weigh1.443× 10−25 kg with s-wave scattering length of 100a0.
these pairs contributes an energy U to the lattice site. This then gives us
En =
n2
U =n(n− 1)
2U. (2.39)
Consider the case of zero tunneling, so that J = 0. The Hamiltonian is then
simplified to
H =1
2U∑i
ni(ni − 1)−∑i
µini. (2.40)
It is clear that the eigenstate is the one where each lattice site has an integer number
of atoms ni. Furthermore, the Hamiltonian is now an independent sum of all of the
lattice sites. Therefore for the case of J = 0, we need only consider a single lattice
32
site. The energy of a single lattice site with n atoms is then
En =n(n− 1)
2U − µn. (2.41)
The energy of a lattice site with n− 1 atoms is
En−1 =(n− 1)(n− 2)
2U − µ(n− 1). (2.42)
Consider the energy difference between the configuration with n − 1 atoms and n
atoms. Since the two terms in the energy have opposite signs, it is possible for there
to be a critical n where it is just barely favorable to have one more particle in the
site. The relevant energy to determine this critical n is the energy different between
En and En−1. This is found to be
∆En ≡ En − En−1 = (n− 1)U − µ. (2.43)
The maximum possible n is determined by ∆En < 0. This implies
nmax <µ
U+ 1⇒ nmax = Int
( µU
)+ 1. (2.44)
This discreteness of the possible occupation number is a distinct signature of the
Mott insulator state and is what gives rise to the so-called ‘wedding cake structure.’
Such a discrete spatial structure is particularly amenable to the technique of Bragg
scattering, which is described in detail in this thesis [77].
2.3.1 Mott Insulator Phase Diagram
Depending on the parameter J/U and µ/U , the system can be either in a Mott
insulator or superfluid phase [79, 108]. The phase diagram can be found by considering
the energy of the system using second-order perturbation theory. This requires a
mean-field approximation, where the nearest-neighbor interactions as expressed by
the term in the Hamiltonian with J are decoupled into single-site operators and an
33
order parameter ψ, which in this case is ψ =√n. Furthermore, Landau’s theory of
second order phase transitions says that phase transitions occur when d2E/dψ2 = 0,
where the energy E can be expressed as E = a0 + a2ψ2 +O(ψ4), which requires that
at the phase transition, a2 = 0. With this understanding, the phase boundary is
given by the equation
U
zJ=
n+ 1
n− µ/U +n
1− n+ µ/U, (2.45)
where z is the coordination number (2× 3 = 6 in the case of a 3 dimensional lattice),
and the occupation number n is given by n = Integer(µ/U) + 1 for µ/U > 0 and
n = 0 otherwise. The phase diagram determined by this expression is given in Fig. 2-
4, where regions of occupation number up to 5 are shown.
0 1 2 3 4 50
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2Phase Diagram of the Mott Insulator
µ/U
zJ/U
n=1n=2n=3n=4n=5
Figure 2-4: Phase diagram of the superfluid to Mott insulator transition. The coloredregions correspond to the Mott insulator phase, and the white regions correspond tothe superfluid phase. Occupation number regions of up to 5 are shown. The boundaryis given by Eq. 2.45.
34
2.3.2 Non-Zero Temperature Effects
So far our description of the Bose-Hubbard Hamiltonian has assumed zero temper-
ature. But of course in reality the system realized in an experiment is at a non-zero
temperature. Understanding the system at non-zero temperature is particularly im-
portant when one is hoping to realize even lower energy Hamiltonians such as the
Heisenberg Hamiltonian which we describe in the next section. Our results on spin
gradient thermometry and demagnetization cooling, included in this thesis as appen-
dices D, E, and F, relied strongly on our understanding of non-zero temperature
effects.
Consider a Mott domain with n atoms per site. Then it is reasonable to expect that
the lowest-lying excited states are particle and hole states, where an additional particle
has been added/removed from the background occupation number of n [36, 51]. The
energies of a particle and a hole are
En,particle =(n+ 1)n
2U − n(n− 1)
2U = nU, (2.46)
En,hole =(n− 1)(n− 2)
2U − n(n− 1)
2U = −(n− 1)U. (2.47)
This particle-hole approximation allows us to truncate the Hilbert space of a lattice
site to states with n− 1, n, and n+ 1 atoms. Other states are suppressed by factors
of e−βU . Assuming βU 1, this approximation is expected to be valid. In general,
the partition function for a single site can be written,
z =∑n
e−βH(n) =∑n
e−β(n(n−1)U/2−µn). (2.48)
Truncating this to the relevant states and factoring the common term, we get
z = e−βH(n−1) + e−βH(n) + e−βH(n+1) (2.49)
∝ e−β(−(n−1)U+µ) + 1 + e−β(nU−µ) (2.50)
= e−β(Ehole+µ) + 1 + e−β(Eparticle−µ) (2.51)
35
With this partition function, we can write down the probability for there to be a hole
and particle in any given Mott region with occupation number n. This is found to be
phole =e−β(Ehole+µ)
z(2.52)
pparticle =e−β(Eparticle−µ)
z. (2.53)
We can now calculate the mean occupation number at non-zero temperature as a
function of the chemical potential and the interaction energy, given the probability
of holes and particles. This is given by
n = n+ pparticle − phole. (2.54)
This is plotted as a function of µ/U for various values of U/kBT in Fig. 2-5. One can
see that the sharp steps a T = 0 become rounded as the temperature increases, and
roughly at kBT/U ≈ 0.2, the sharpness has almost completely faded away. This value
of 0.2 is commonly accepted as the ‘melting temperature’ of the Mott insulator.
2.4 Heisenberg Hamiltonian and Quantum Mag-
netism
The Heisenberg Hamiltonian is a model which describes spin-spin interactions
in a lattice and gives rise to phases such as ferromagnetic and antiferromagnetic
states [29]. This Hamiltonian can be realized from the Bose-Hubbard Hamiltonian in
the perturbative limit of J/U 1, where the relevant energy scale becomes J2/U , also
called the superexchange energy scale. This energy scale corresponds to a temperature
of order 100 pK for rubidium atoms in a 3D lattice of lattice spacing a few hundred
nanometers. Our work on spin gradient thermometry and demagnetization cooling,
included in this thesis as appendices D, E, and F, were particularly driven by our
goal of realizing the Heisenberg Hamiltonian which requires very low temperatures.
Our work on Bragg scattering, described in chapter 3, was undertaken with the goal
36
−1 −0.5 0 0.5 1 1.5 20
0.5
1
1.5
2
2.5
3Average Occupation Number as a Function of Chemical Potential
µ/U
n aver
age
kBT/U=0
kBT/U=0.01
kBT/U=0.05
kBT/U=0.1
kBT/U=0.2
Figure 2-5: Average occupation number as a function of chemical potential for varioustemperatures. These steps are often call the ‘wedding cake structure.’
of applying it to study spin ordering once the Heisenberg Hamiltonian is realized.
Let us start with a system where two different effective spin species interact dif-
ferently in a lattice [29]. Denote the two relevant internal states with the effective
spin index σ =↑, ↓ respectively. These two spins are trapped in a perodic potential
Vµσ sin2(kµ · r) in a certain direction µ where kµ is the wavevector of light. For suffi-
ciently strong periodic potential and low temperatures, the low energy Hamiltonian
is given by
H = −∑〈i,j〉,σ
(tµσa
†iσajσ + H.c.
)+
1
2
∑i,σ
Uσniσ (niσ − 1) + U↑↓∑i
ni↑ni↓, (2.55)
where 〈i, j〉 denotes the nearest neighbor sites in the direction µ, aiσ are bosonic
annihilation operators for atoms of spin σ localized on site i, and niσ = a†iσaiσ.
Using the harmonic Wannier functions, the spin-dependent interaction energy U↑↓
37
is given by
U↑↓ =4πa↑↓~2
m
∫d3x |w↑(x)|2 |w↓(x)|2 (2.56)
≈√
8
πka↑↓
(ERV1↑↓V2↑↓V3↑↓
)1/4, (2.57)
where Vµ↑↓ = 4Vµ↑Vµ↓/(V1/2µ↑ + V
1/2µ↓ )2 is the spin average potential in each direction.
2.4.1 Anisotropic Heisenberg Hamiltonian
Using a generalization of the Schrieffer-Wolff transformation [50, 89] (or another
method [67]), to leading order in tµσ/U↑↓, Eq. 2.55 is equivalent to the effective Hamil-
tonian
H =∑〈i,j〉
[λµzσ
zi σ
zj − λµ⊥
(σxi σ
xj + σyi σ
yj
)], (2.58)
where σzi = ni↑ − ni↓, σxi = a†i↑ai↓ + a†i↓ai↑, and σyi = −i(a†i↑ai↓ − a†i↓ai↑) are the usual
spin operators. The parameters λµz and λµ⊥ are given by
λµz =t2µ↑ + t2µ↓
2U↑↓−t2µ↑U↑−t2µ↓U↓
and λµ⊥ =tµ↑tµ↓U↑↓
. (2.59)
Note that the Hamiltonian in Eq. 2.58 is the well-known anisotropic Heisenberg model
(XXZ model) which arises in various condensed matter systems [10].
One can also consider the case where an external magnetic field along the z di-
rection is applied. The Hamiltonian then becomes
H =∑〈i,j〉
[λµzσ
zi σ
zj − λµ⊥
(σxi σ
xj + σyi σ
yj
)]−∑i
hzσzi . (2.60)
Now assume that the tunneling and same-spin interaction energies are the same for
the two spin states. We can perform a variational calculation on the minimum energy
of this system by two trial functions (i) a Neel state in the z direction 〈σi〉 = (−1)iez;
(ii) a canted phase with ferromagnetic order in the x−y plane and finite polarization
38
in the z direction 〈σi〉 = cos θex + sin θez. Here θ is a varational parameter and ex,z
are unit vectors in the direction x, z. We then take the expectation value of H for
each case to get
〈H〉N = −Zλz2, and 〈H〉F =
Zλz2
sin2 θ − Zλ⊥2
cos2 θ − hz sin θ, (2.61)
where 〈H〉N and 〈H〉F are the Neel and canted phases respectively and Z is the
coordination number corresponding to the number of nearest neighbors.
We can minimize the expression for the canted phase with respect to θ to get
cos θ(Z(λz + λ⊥) sin θ − hz) = 0. (2.62)
The energy is minimized when
cos θ = 0 or sin θ =hzZλ⊥
× 1
1 + λz/λ⊥. (2.63)
The two values that minimize the energy are θ = π/2, sin−1(hz/Z(λz + λ⊥)). Note
θ = π/2 corresponds to the z ferromagnetic phase, and whose energy is given by
〈H〉z =Zλz
2− hz. (2.64)
At this point, we are ready to determine the phase boundaries between the z
ferromagnet, xy ferromagnet, and the antiferromagnet states. We can let y ≡ λz/λ⊥
and x ≡ hz/Zλ⊥ for ease of manipulation. Then if we consider 〈H〉z = 〈H〉F, we
obtain
y = x− 1⇒ hz = Z (λz + λ⊥) . (2.65)
We can also consider 〈H〉N = 〈H〉F to obtain
y =√x2 + 1⇒ hz = Z
√λ2z − λ2
⊥. (2.66)
39
The phase diagram dictated by Eqs. 2.63, 2.65 and 2.66 is given in the left plot
of Fig. 2-6. For an actual ultracold atomic experiment, we can assume that there are
no spin-flips and that the density of atoms throughout the sample must remain the
same. To explore this phase space, we can apply a controlled magnetic field gradient,
so that B ≈ x(dB/dx) and can take a slice along the horizontal direction with one
sample. If we create a sample with an equal mixture of ↑ and ↓ atoms, then from the
previous assumption, the only thing that the spins can do is rearrange themselves in
the sample. Spin ↑ atoms will congregate in the lower field region and spin ↓ atoms
will congregate in the higher field region. Then it is natural to expect the center of
the sample to have zero spin, and the sample will have a symmetric spin profile about
the center.
As can be seen in the Hamiltonian, the energy scale for these phases are of or-
der t2/U , which for typical experimental paramters for 87Rb are on the order of 100
pK [15]. Thus this low energy scale causes difficulties in preparing such phases of mat-
ter. However, recent work performed in this group shows that such low temperatures
can be achieved by the method of adiabatic spin gradient cooling [76].
These phase diagrams represent magnetic phases of matter, which in its fermionic
incarnation is believed to form the basis for high-temperature superconductors [69].
Thus there is great technological as well as scientific interest in understanding the
properties of this Hamiltonian.
2.4.2 Quantum Magnetism and Bragg Scattering
The Bose-Hubbard Hamiltonian has been studied extensively since its observation
in 2002 [42]. One major step in this field would be to study an even lower temperature
phase, which is the Heisenberg Hamiltonian. Bragg scattering will be a powerful tool
to study the magnetic phases of matter arising from the Heisenberg Hamiltonian
by making use of spin-dependent scattering, as was done in neutron scattering to
observe the antiferromagnetic state [91]. Chapter 3 describes our work to study the
Bose-Hubbard Hamiltonian with Bragg scattering, thus paving the way towards its
employment in the search for quantum magnetism in optical lattices.
40
0 0.5 1 1.5 2 2.5 3−2
−1.5
−1
−0.5
0
0.5
1
1.5
2
2.5
3Phase Diagram of the 2−component Spin Hamiltonian
hz/Zλ⊥
λ z/λ⊥
I
II
III
−3−2
−10
12
3
−2
−1
0
1−1
−0.5
0
0.5
1
hz/Zλ⊥
Phase Diagram of the 2−component Spin Hamiltonian
λz/λ⊥
Sz
Figure 2-6: Phase diagrams of the Heisenberg Hamiltonian. The left figure is charac-terized by the following order parameter: I, z-Neel order; II, z-ferromagnetic order;and III, xy-ferromagnetic order. The right figure is the same plot as the left figurebut with the spin along the magnetic field direction (sin θ) represented by the verticalaxis.
2.5 Harper Hamiltonian and Gauge Fields
Gauge fields play an important role in various areas of modern physics, from
particle physics, where the Standard Model relies heavily on gauge fields [81], to
condensed matter physics, where gauge fields can explain the integer and fractional
quantum Hall effects [107, 68]. Quantum Hall effects are interesting because they
can manifest topological properties of wavefunctions, which could form the basis
of a robust quantum computing platform due to the fact that environmental noise
have a much more difficult time perturbing topological properties of matter [64].
In condensed matter, gauge fields can help to explain the dynamics of electrons in
magnetic fields. However, gauge fields in condensed matter systems can be more
general than magnetic fields because they can describe non-abelian gauge fields [94,
48].
One particular type of Hamiltonian that can give rise to such quantum Hall ef-
fects is the Harper Hamiltonian [105]. This Hamiltonian describes the dynamics and
energy levels of electrons moving in the presence of a magnetic field and a background
lattice in the tight-binding limit [49]. This Hamiltonian has been shown to exhibit
41
Figure 2-7: The Hofstadter butterfly [52]. This plot shows the fractal nature of theHarper Hamiltonian. The x-axis is in units of the eigenenergy per tunneling energyand the y-axis is in units of the enclosed flux quanta.
a fascinating fractal band structure which is colloqually called the ‘Hofstadter but-
terfly’, named after the discover Douglas Hofstadter [52]. Such a band structure is
shown in Fig. 2-7.
The particular Harper Hamiltonian we are interested in can be written as
H = −K∑m,n
(eiφm,na†m+1,nam,n + e−iφm,na†m,nam+1,n
)− J
∑m,n
(a†m,n+1am,n + a†m,nam,n+1
)(2.67)
where K and J are tunneling amplitudes in the x and y directions respectively, a†m,n
and am,n are the creation and annihilation operators, respectively, at site (m,n), and
the indices m and n are for the x and y lattice positions respectively. In general, for
the scheme we are interested in we can write φm,n = mφx+nφy. In chapter 4 I describe
in more detail how this Hamiltonian can be engineered with ultracold atoms, but for
the purposes of this section I would like to focus on the properties of this Hamiltonian.
42
2.5.1 Electromagnetism, Gauge Fields, and Wavefunctions
The phase that the wavefunction acquires upon tunneling is directly related to
another important concept in physics, namely that of gauges [86, 44, 109].
In electromagnetism, the electric field E and magnetic field B can be written in
terms of the scalar potential φ and vector potential A as
E = −∇φ− ∂A
∂tand B = ∇×A. (2.68)
Quantum mechanically, the Hamiltonian can be written in the following way to in-
corporate the scalar and vector potentials
H =(p− qA)2
2m+ qφ. (2.69)
Now, it is possible to rewrite φ and A in a way such that it leaves the electric and
magnetic fields unchanged. Namely,
φ′ = φ− ∂Λ
∂tand A′ = A +∇Λ, (2.70)
where Λ is in general any function of space and time. Thus the particular form of the
scalar and vector potentials that give rise to specific electric and magnetic fields are
not unique, but is defined up to a gauge given by the function Λ [57]. One speaks
of ‘choosing a gauge’ in solving specific problems. For example, one could choose the
gauge ∇ ·A = 0.
Here let us consider the situation of time-independent fields and potentials. If you
have a particle moving in the presence of some vector potential A, the wavefunction
of the particle evolves as [2]
ψ = ψ0 exp
[iq
~
∫A · dx.
](2.71)
Now we will consider a specific example where this phase of the wavefunction can
manifest itself in an observable.
43
Source of
Electrons
Solenoid Screen
Path 1
Path 2
B !
B=!
Figure 2-8: Setup for the Aharonov-Bohm effect. Electrons emerge from the electronsource to the left, traverse through two slits, and then impinge on a screen to theright. In between is an infinitely long solenoid which has a non-zero magnetic fieldinside, but zero magnetic field outside. Even though the magnetic field is zero wherethe electrons are, they feel the effect of the magnetic field inside the solenoid.
Consider the following set up, shown in Fig. 2-8. We have a source of electrons
to the left, which pass through two slits. Beyond the slits, there is an infinitely long
solenoid pointing out of the page which is also impenetrable to the electrons. Thus
we can consider this solenoid to have zero magnetic field anywhere outside of the
solenoid. Note however that the vector potential can still be non-zero in the outer
region. After passing through the slits the electrons impinge on a screen.
Now if we consider the wavefunction of an electron which traveled around path 1
and path 2 to the same point on the screen, the probability to observe the electron
will be determined by the sum of the two wavefunctions. This can be written
ψtot = ψ1 exp
(iq∫
path1dx ·A
~
)+ ψ2 exp
(iq∫
path2dx ·A
~
)(2.72)
= exp
(iq∫
path2dx ·A
~
)[ψ1 exp
(iq∮dx ·A~
)+ ψ2
](2.73)
= exp
(iq∫
path2dx ·A
~
)[ψ1 exp
(iqΦ
~
)+ ψ2
], (2.74)
where we used the fact that the integral∮dx·A is exactly the magnetic flux Φ through
the solenoid. The observable is the wavefunction squared, and due to the complex
44
phase, the resulting probability of observing an electron on the screen will oscillate
as a function of the magnetic flux. It is very curious that even though the electron
never enters the region with non-zero magnetic field, the probability to observe an
electron depends on the magnetic field in the forbidden region. This is the celebrated
Aharonov-Bohm effect [2], also sometimes called the Ehrenberg-Siday effect [30]. This
effect beautifully illustrates the fact that two different paths can imprint non-trivial
topological phases on wavefunctions just by enclosing a region with magnetic flux.
2.5.2 Flux Density per Plaquette α
Now that we have an appreciation of the connection of topological properties to
the phase of the wavefunction, which in turn can be understood as a particular gauge
chosen for the system, let us return to the Harper Hamiltonian written as Eq. 2.67 and
understand how the phase relates to an effective magnetic field. As we saw previously,
we can write the phase as φm,n = mφx + nφy where m and n are integers describing
the position in the two-dimensional lattice and φx,y are constants. Then a phase will
be accumulated upon tunneling in the x direction. In this case the phase accumulated
can be shown as in the left of Fig. 2-9. We see that the total phase accumulated if
one atom tunnels around one full square, called a plaquette or a unit cell, is Φ = φy.
Note that the total accumulated phase only depends on the y dependence of the phase
accumulated upon tunneling in the x direction.
The flux density per plaquette is defined as α ≡ Φ/(2π). This parameter has a
strong connection to electrons with charge e in a square two-dimensional lattice of
side a in a magnetic field B [52]. Hofstadter defines this parameter as (in SI units)
α =a2B
h/e. (2.75)
This is precisely the magnetic flux enclosed by a square of side a divided by the
magnetic flux quantum of an electron. This shows that for a typical crystal with lattice
spacing of a few angstroms, magnetic fields on the order of a billion gauss is necessary
to achieve α of order unity. However, very recently researchers have devised a clever
45
m
n
3
2
1
2 1 3
x
y
yx !
yx !2
yx 2
yx!! 22
yx!! 3
yx!! 32
(a)
m
n
3
2
1
2 1 3
x
y
y
y
y
y
(b)
Figure 2-9: Phase accumulated upon tunneling and enclosed flux. (a) shows the phaseaccumulated upon tunneling from site (m,n) to (m+ 1, n). (b) shows the total phaseaccumulated after tunneling as (m,n)→ (m,n+ 1)→ (m+ 1, n+ 1)→ (m+ 1, n)→(m,n), which is φy for each such plaquette.
scheme by using boron nitride and a layer of graphene on top and slightly rotated
from each other to create an effectively much more widely spaced superlattice on the
order of tens of nanometers. This allowed the researchers to observe the signatures
of the Hofstadter butterfly through conductivity measurements in magnetic fields of
tens of tesla [26, 54]. Our proposal with ultracold atoms in optical lattices allow us
to achieve effectively high magnetic fields by making use of Raman transitions which
impart phases upon tunneling, which will be discussed in more detail in chapter 4.
2.5.3 Band Structure for α = 1/2
As an example, here we we consider the specific case φx = −π and φy = π which
leads to φ = π(−m + n) and therefore α = 1/2. This is the particular type of phase
we initially study experimentally, but any other phase dependence and consequently
any other α is also realizable. Schematically the Hamiltonian can be represented as
the left part of Fig. 2-10.
The 2D square lattice can be divided into two sub-lattices, where an atom on one
sub-lattice accumulates a phase of 0 when it tunnels one site in the +x direction, and
an atom in the other sub-lattice accumulates a phase of π when it tunnels one site in
the +x direction. These two sub-lattices are represented by filled and unfilled circles
46
m
n
3
2
1
2 1 3
x
y
0
0
0
xk
yk
a
k Space
Brillouin
Zone
Figure 2-10: Schematic of a two-dimensional lattice with phases. The left figureis in real space, where the phase accumulated upon tunneling is represented. Theright figure is the reciprocal space representation, along with the first Brillouin zonehighlighted.
in the left of Fig. 2-10. With these two sub-lattices, we can define primitive vectors
and basis vectors [9].
We see that the primitive vectors are a1 = a(x− y) and a2 = a(x + y) where a is
the lattice spacing. The basis vectors are b1 = 0 and b2 = ax. The reciprocal lattice
vectors are c1 = πa(x − y) and c2 = π
a(x + y). The Brillouin zone is then a tilted
square at θ = 45 and is shown in the right of Fig. 2-10. With this information, we
can calculate the band structure for this system.
The Hamiltonian in matrix form is,
H =
0 J(eikya + e−ikya
)+K
(eikxa − e−ikxa
)J(e−ikya + eikya
)+K
(e−ikxa − eikxa
)0
(2.76)
=
0 2J cos(kya) + 2iK sin(kxa)
2J cos(kya)− 2iK sin(kxa) 0
. (2.77)
The band structure then is
E±(kx, ky) = ±2√J2 cos2(kya) +K2 sin2(kxa). (2.78)
47
kx
k y
−1 −0.5 0 0.5 1−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
Figure 2-11: Band structure of the α = 1/2 Harper Hamiltonian. It is a contour plotof the lowest band structure, where we have two minima and two Dirac cones in theBrilloin zone, the boundary which is given in red bold lines.
The lower branch is plotted in Fig. 2-11. We see that the band minima are at
(kx, ky) = (±π/(2a), 0). Furthermore, the band has Dirac points at (kx, ky) =
(0,±π/(2a)) where the band gap between the two bands is zero.
Another approach to solving the same problem is to use the concept of magnetic
unit cells [121, 122, 78, 83]. This approach is well-suited to lattice systems which have
such non-trivial phases associated with tunneling between neighboring lattice sites.
Different phase dependencies corresponding to different α will give rise to different
band structures. For non-rational α, the system will not have any periodicity and
so a band description will not be appropriate, but practically any α realized will
have some spread and will include rational α and so a band description should be
adepquate even if one does not precisely achieve rational α [52]. Furthermore, if
interactions are added to the Harper Hamiltonian, even more fascinating states such
as fractional quantum Hall states are expected to arise [94, 48]. Therefore the Harper
Hamiltonian provides an exciting new frontier in the study of topological states of
matter, both with and without interactions.
48
Chapter 3
Bragg Scattering as a Probe of
Ordering and Quantum Dynamics
This chapter elaborates on the work published as Bragg Scattering as a Probe of
Atomic Wave Functions and Quantum Phase Transitions in Optical Lattices, Phys.
Rev. Lett. 107, 175302 (2011) [77].
As discussed in the introduction, ultracold atomic gases are an ideal system in
which to study many-body phenomena because of the relative ease with which pa-
rameters in the model Hamiltonian can be tuned across a wide range [70, 13]. Such
studies have resulted in a better understanding of various phase transitions such as
the Berezinskii-Kosterlitz-Thouless transition in two dimensional systems [47], the
BEC-BCS crossover of interacting fermions [123] and the superfluid to Mott insulator
transition in a three-dimensional lattice [42]. One major goal of this field is to realize
spin phases such as antiferromagnetic states to explore quantum magnetism and its
interplay with high-temperature superconductivity [69].
Concurrently, there are numerous efforts to develop techniques to probe and un-
derstand the atomic ensemble once a new type of ordering is achieved. One recent
development is the realization of single-site resolution of atoms in two-dimensional
optical lattices [11, 90]. An alternative method to measure in situ spatial ordering
is the technique of Bragg scattering, often used in a condensed matter context to
determine crystal structure [9]. In particular, Bragg scattering with neutrons led to
49
the discovery of antiferromagnetism in the solid state [91] and with x-rays led to the
discovery of the double helix structure of DNA [111].
Bragg scattering relies on the interference of waves scattered coherently from an
ensemble of particles. In particular when atoms are arranged in a periodic pattern
in three spatial dimensions, there are certain angles of the incoming and reflected
beams where scattering is dramatically enhanced compared to other angles. This has
allowed crystallographers to use x-rays to determine the properties of crystals such as
lattice geometry at the angstrom scale. We have applied this technique to ultracold
atoms in a three-dimensional optical lattice by scattering near-infrared light where
the atoms are spaced almost 104 times farther apart than those in typical condensed
matter samples.
Pioneering works on Bragg scattering from cold atoms in optical lattices were
done by the Hansch and Phillips groups using laser cooled atoms [113, 12, 84, 112].
These lattices were sparsely populated and the atoms occupied several bands. In
this Letter, we have used Bragg scattering to study bosonic atoms cooled to quantum
degeneracy and placed in a three dimensional cubic lattice where the atoms occupy the
lowest band. In particular, we have measured directly the Heisenberg-limited width
of both position and momentum of the single ground state atomic wavefunction in an
optical lattice. Furthermore, there is a revival of Bragg scattered light some time after
releasing the atoms from the optical lattice, analogous to the optical Talbot effect.
This signal gives a direct measure of the coherence of the superfluid component in
the lattice.
3.1 Experimental Setup for Bragg Scattering
Laser cooling and evaporative cooling are used to achieve quantum degeneracy
of 87Rb atoms in the |F = 2,mF = −2〉 state [100] which are loaded into a crossed
dipole trap whose trap frequencies range between 30 and 160 Hz. Once quantum
degeneracy is achieved, the optical lattices generated by a single 1064 nm fiber laser
are adiabatically ramped on. The lattices are calibrated by applying a 12.5 µs pulse
50
Probe in Bragg out
[1 0 0] Planes
43
x
y
z
Figure 3-1: Bragg scattering schematic setup. The atoms are arranged in a cubiclattice structure with spacing of 532 nm, and the probe beam is a laser at a wavelengthof 780 nm, close to one of the rubidium transition lines. The thin red lines indicatethe [1 0 0] planes of the lattice.
and measuring the Kapitza-Dirac diffraction of the atoms and comparing this to
theory. The system typically contains about 105 atoms, which leads to up to 5 atoms
per lattice site. The Bragg reflected light is detected on a CCD camera which images
along a direction which satisifies the Bragg condition.
The Bragg scattering condition for a three-dimensional cubic lattice is given by
2d sin θ = nλp (3.1)
where d is the spacing between lattice planes, θ is the angle between the incoming
beam and the lattice planes, n is any positive integer and λp is the wavelength of the
probe beam. For our experiment λp is 780 nm and d is 532 nm. With these conditions
the only allowed angle θ is 47 where n is one and corresponds to Bragg reflection off
the [1 0 0] plane or any equivalent plane. A schematic of the probe beam with respect
to the atoms in the lattice is shown in Fig. 3-1. Since the full angular width of the
Bragg scattered light is small (measured to be 4.1± 0.4), a precise alignment of the
incident beam had to be performed at the 1/min repetition rate of the experiment. In
contrast, in one and two dimensions, diffractive light scattering occurs at any angle
of incidence, as recently shown with atoms in a two-dimensional optical lattice [114].
51
The probe beam at the Bragg angle had a typical power of 0.3 mW and beam
diameter of 300 µm, large enough to illuminate the entire atomic cloud, and was
detuned from the 52S1/2, F = 2→ 52P3/2, F′ = 3 cycling transition of 87Rb by a few
natural linewidths, where the natural linewidth Γ is 6 MHz. The detuning needs to
be sufficient so that the light traverses the entire atom cloud. Although the absolute
signal varies with the detuning, the conclusions reached in this Letter appear not
to depend strongly on the amplitude or sign of the detuning. The probe beam was
applied for a few microseconds, enough to obtain a signal but short enough to avoid
heating effects.
In the following subsection I describe how we characterized and determined the
optimum parameters for the Bragg beam.
3.1.1 Characterization of Bragg Scattering
In this subsection I elaborate on how we characterized our Bragg scattered light
so that we obtained the optimum signal to study atomic dynamics and many-body
effects.
A schematic of the optics setup for the Bragg beam is shown in the top of Fig. 3-2.
We use a guide mirror at the focal point of our telescope system, which adjusts the
incident angle of the Bragg probe upon our atom cloud while keeping the beam aligned
on the atoms at all angles. The exact angle required to achieve Bragg scattering
between the incident probe beam and the atoms in the lattice is determined by the
Bragg condition, but in reality due to the finite size of the sample, there is a range
of angles that will still lead to Bragg scattering. Thus we characterized the range of
angles that will give Bragg scattering and optimized our alignment to maximize our
signal. We accomplished this by scanning the guide mirror angles.
The Bragg scattered light as a function of the vertical and horizontal alignment of
the guide mirror are given in the bottom of Fig. 3-2. The mirror mount is a Thorlabs
KM200, which has 1/4”-80 adjuster screws with a resolution of 5 mrad (0.3) per
revolution. If we use the full width half maximum (FWHM) of our signal with −500
counts being the background, the vertical FWHM is 1150 and the horizontal FWHM
52
f = 500 mm f = 400 mm
500 mm 900 mm 400 mm
Guide
Mirror
Collimated Bragg Beam
Atoms
f = 400 mm
400 mm
−1500 −1000 −500 0 500 1000−500
0
500
1000
1500
2000
2500
3000
3500
Fco
unt
Vertical Knob Angle (Degrees)−400 −200 0 200 400 600
−500
0
500
1000
1500
2000
2500
3000
3500
Fco
unt
Figure 3-2: Bragg reflection as a function of the guiding mirror screw angle. Top:schematic for the optics setup for our Bragg beam. Bottom: plots of Bragg scatteredlight vs. the vertical (left) and horizontal (right) alignment of the guide mirror.With the known screw specifications, these correspond to a FWHM angular extentof 0.9 and 0.3 respectively. This agrees well with simple estimates of these valuesas described in the text.
is 400 in adjuster screw angles, which correspond to 0.9 and 0.3 respectively in
true beam angles. Note the relationship θ ∼ λ/2d, where θ is the angular extent of
the cloud, λ is the laser wavelength, and d is the spatial extent of the cloud [113].
From this we have θH/θV = dV /dH , where θH and θV are the angular extent in the
horizontal and vertical directions respectively and dH and dV are the spatial extent in
the horizontal and vertical directions respectively. From our trap geometry, we expect
this to be dV /dH = 1/3, and experimentally we find that θH/θV = 0.3/0.9 = 1/3,
giving good agreement.
The detuning of the Bragg beam from the atomic resonance is an important
53
parameter. If the beam is too close to resonance, then the atom cloud will be opaque
to the beam, and the entire cloud cannot coherently scatter at the same time and
simply lead to fluorescence as well as re-absorption by other atoms. On the other
hand, if the beam is detuned too far, the atoms will not interact sufficiently strongly
and will not lead to any scattering, coherent or incoherent. Thus the detuning must be
chosen far enough so that the Bragg beam penetrates the entire cloud, but not so far
that the atoms do not scatter much of the Bragg beam. We determined experimentally
that we see no Bragg scattering lower than 4Γ detuning, where Γ = 6 MHz is the
natural linewidth of the rubidium D2 line. At 8Γ, we see Bragg reflection but little
fluorescence. Therefore the optimum initial setting for searching for Bragg reflection
is 6Γ detuning, so that we have fluorescence to guide the eye, but the Bragg signal is
observable if the Bragg condition is satisfied.
The optical power in the Bragg beam is another important parameter for much
the same reason as how the detuning is important if we are to obtain the strongest
signal in the shortest amount of time possible. For example, if the power is too low,
not enough photons are scattered. Thus we measured the Bragg reflection by varying
the power in the Bragg beam, and a representative response is shown in Fig. 3-3,
where the probe was on for 3 µs and the detuning was 8Γ. We see that the optimum
power for observing the maximum Bragg reflection in this case is at around 0.8 mW.
The duration that the Bragg probe is turned on is also a relevant parameter to
obtain the best signal. The longer the beam is on, the more Bragg scattering will
happen, until the atoms have scattered enough photons to be ejected from the ground
state of the lattice at which point all of the Bragg scattered light obtainable has been
obtained. This plateau behavior is shown in Fig. 3-4. We see that the optimum probe
time is at about 5 µs where the Bragg signal reaches a limit.
We also determined the spatial extent of the Bragg scattered beam. This was done
by positioning a razor blade perpendicular to the beam path and partially blocking
the beam. This works because the Bragg reflection is expected to produce a fairly
well collimated beam. The schematic of the setup and data are shown in Fig. 3-5.
From the spatial extent of the beam, we can deduce the angular width of the beam
54
0 0.5 1 1.5 20
500
1000
1500
2000
2500
Fco
unt
Bragg Probe Power (mW)
Figure 3-3: Bragg scattering as a function of Bragg probe power. For this case of 8Γdetuning, the maximum Bragg reflection is at about 0.8 mW.
0 2 4 6 8 10−1000
0
1000
2000
3000
4000
Fco
unt
Probe Time (µs)
Figure 3-4: Bragg scattering as a function of Bragg probe time. For this case of 8Γdetuning, we see that the Bragg signal reaches a plateau at about 5 µs.
55
coming from the atom cloud. A fit of the data to an error function gives the width
to be 4.5 mm and using the fact that the imaging lens has f = 125 mm, the half
angular size of the beam is 2.0. This is in rough agreement with our estimate given
by θ ∼ λ/2d where λ is the wavelength of the Bragg probe and d is the spatial extent
of the cloud [113]. If we take the cloud size to be 10 µm and the λ = 780 nm, we
have θ ∼ 0.039 radians, or 2.2, which is in rough agreement with our measurement.
With the optimum parameters that maximize our signal from the Bragg beam,
we have studied properties of atoms in a three-dimensional optical lattice.
3.2 Debye-Waller Factor and Bragg Scattering
As the lattice is increased the Bragg reflected signal increases as expected from
the crystal ordering and tighter localization of the atoms (Fig. 3-6). For this data,
the lattice in the z direction, which is also the direction of gravity, was ramped to
15ER, where ER = h2/(2mλ2L) is the recoil energy, h is the Planck constant, m is the
mass of 87Rb and λL = 1064 nm is the lattice laser wavelength. Simultaneously the
lattices in the horizontal directions were ramped to various lattice depths. Note that
for our geometry the Bragg reflected intensity is insensitive to the lattice depth in
the z direction because Bragg scattering occurs only in the horizontal xy plane, as
we discuss later.
The scattering of light by a collection of atoms can be modeled in the following
way. In the limit of low probe intensity and low optical density of the cloud, the Born
approximation can be used. Then the scattering cross-section dσ/dΩ can be written
as a product of one-dimensional Debye-Waller factors where
dσ
dΩ∝∏i
exp
[−(∆ri)
2K2i
2
], (3.2)
where i = x, y, z is the index for the three dimensions, Ki is the magnitude of the
i-th component of K where K = kin − kout is the difference between the incoming
probe beam and Bragg scattered wavevectors, and ∆ri =√
~/(mωi) is the harmonic
56
Razor distance
Atoms
Razor
Bragg reflection Camera
125 mm
0 5 10 15 20 25−500
0
500
1000
1500
2000
2500
3000
3500
Razor Distance (mm)
Bra
gg S
catte
ring
(arb
.)
Figure 3-5: Bragg reflection as a function of razor position. Top is the experimentalsetup. Bottom is the Bragg scattered signal as a function of razor position at 8Γdetuning. From the error function fit, the width is about 4.5 mm. With an imaginglens of f = 125 mm, this corresponds to an angular extent of 2.0. This is in roughagreement with an estimate of 2.2.
57
Figure 3-6: Bragg scattering as a probe of the spatial wavefunction width. Theplot shows Bragg scattered intensity vs. lattice depth in units of the recoil energy.The right axis gives the corresponding root-mean-square width of the wavefunctionsquared. The solid line is a no-free-parameter curve given by the Debye-Waller factor.Error bars are statistical.
58
oscillator width where ~ is the reduced Planck constant and ωi is the trap frequency
in the i-th direction, which depends on the optical lattice depth. The atoms are ap-
proximated as gaussian wavefunctions, which is a good approximation for sufficiently
large lattice depths. However, we find that even for low lattice depths where sig-
nificant superfluid components are expected, this approximation describes the Bragg
scattered signal well.
For our experimental conditions, Kz = 0 because Bragg scattering occurs in the
horizontal xy plane. The lattice depths are controlled in a way such that they are the
same in both the x and y directions which leads to ∆x ≡ ∆rx = ∆ry. This simplifies
the scattering cross-section to
dσ
dΩ∝ exp
[−(∆x)2K2
2
]. (3.3)
Thus Bragg scattering allows the study of the spatial extent of the atomic wavefunc-
tion.
The data can be compared to a no-free-parameter theoretical line, assuming non-
interacting atoms. The Bragg scattered intensity B(NL) as a function of lattice depth
NL can be written as
B(NL) ∝ exp
[− λ2
LK2
8π2√NL
]. (3.4)
Both the harmonic oscillator width ∆x, which depends on the mass of the rubidium
atom and trap frequency, and change in wavevector K are known. This theoretical
line and the data is shown in Fig. 3-6 and we see good agreement. Thus the Bragg
scattered light as a function of lattice depth can be well-described by a model where
the atoms are assumed to be non-interacting gaussian atomic wavefunctions whose
spatial width is determined by the lattice depth.
59
Figure 3-7: Bragg scattering as a probe of the momentum wavefunction width. Theplot shows Bragg scattered intensity vs. the free-expansion time of the atoms afterrapidly turning off the lattices from 15ER. The decay in signal indicates a meltingof the crystal structure, or in other words spreading of the atomic wavepacket with agiven momentum uncertainty. The gray area is a no-free-parameter curve using theDebye-Waller factor taking into account the probe pulse duration of 5 µs. Error barsare statistical.
3.3 Heisenberg-Limited Wavefunction Dynamics
Bragg scattering can also be used to probe the momentum width of the atomic
wavefunctions. This is done by measuring Bragg reflection as a function of the expan-
sion time of the atoms between a rapid lattice turn off and Bragg probe. In particular,
the two horizontal lattices were turned off from 15ER to 0ER in less than 1 µs and the
lattice in the z direction was kept at 15ER. Turning off the lattices allows the atomic
wavepackets in each individual lattice well to expand freely in those directions. The
data in Fig. 3-7 shows that the signal has decayed in 30 µs. The time it takes to
lose crystal structure is roughly the time it takes for the atoms to move over half of
a lattice distance.
The Debye-Waller factor can be used to determine more precisely how the Bragg
60
reflection should behave as a function of the expansion time with no free parameters,
assuming gaussian atomic wavepackets. This makes use of the well-known time-
dependent behavior of a Heisenberg-limited gaussian wavepacket in free space which
can be written as
(∆x(t))2 = (∆x)2 +(∆p)2t2
m2(3.5)
where ∆x and ∆p are the uncertainty of position and momentum respectively at the
initial time and t is the expansion time [87]. In a previous paper where the atoms were
not quantum degenerate, the momentum distribution was assumed to be determined
by temperature [113]. The results of our work show that the momentum uncertainty
is Heisenberg-limited, meaning (∆x)2(∆p)2 = ~2/4, so the Bragg scattered light B(t)
as a function of expansion time can be written as
B(t) ∝ exp
[−(∆p)2K2t2
2m2
]. (3.6)
This curve is shown in Fig. 3-7. The curve is broadened by taking into account the
probe beam duration which was 5 µs and shows good agreement with data. This
analysis shows that by releasing the atoms from a lattice, we can directly probe the
in situ time evolution of the ground state atomic wavefunctions with spatial extent
of tens of nanometers. Furthermore, the atomic spatial and momentum widths are
seen to be Heisenberg-limited.
3.4 Superfluid Revivals of Bragg Scattering
Coherent many-body non-equilibrium dynamics can also be studied with Bragg
scattering after a sudden release of the atoms from the optical lattice in three dimen-
sions. At low lattice depths the Bragg scattered signal shows revivals as a function of
expansion time because of the rephasing of the superfluid atomic wavefunctions that
were originally confined in the lattices. Therefore, these revival signals should give a
measure of the superfluid order parameter. Furthermore, the revivals are analogous
61
to the optical Talbot effect, whose atomic version was observed previously for a ther-
mal beam [16] and for a quantum degenerate gas in the temporal domain [28]. Here
we observe the atomic Talbot effect for an interacting quantum degenerate gas in a
three-dimensional optical lattice. Revivals can be seen in Fig. 3-8 and become less
pronounced as the lattice depth increases.
In particular, the characteristic revival time is determined by
τ =h
~2(2k)2
2m
(3.7)
and is 123 µs for our parameters, where 2k corresponds to the wavevector of the
matter wave. However, with Bragg scattering we expect the first revival at half
that time of 61 µs, which is what we observe in Fig. 3-8. This can be understood
by realizing that the atomic wavepackets need only travel half the lattice spacing to
constructively interfere with the wavepackets traveling in the opposite directions from
the nearest neighbor sites.
3.5 Superfluid to Mott Insulator Through Bragg
Scattering
The superfluid coherence as a function of lattice depth can be studied by com-
paring the Bragg reflected signal when the atoms are in the optical lattice to the
signal at the first revival. The revival signal as a function of lattice depth is shown
in Fig. 3-9, where the superfluid to Mott insulator transition is expected to occur
around 13ER [42]. In a non-interacting system without any dissipation, one would
expect complete revivals. On the other hand, in the Mott insulating phase where
phase coherence has been lost, revivals should completely disappear, except for a
very weak signal at intermediate lattice depths due to particle-hole correlations [38].
To understand the measured revival decay, we have numerically studied the one-
dimensional Gross-Pitaevskii equation that assumes interacting matter waves at zero
temperature. The simulations show that interactions and finite size effects have neg-
62
Position
Tim
e
Figure 3-8: Revivals of Bragg scattered light. The plot shows the Bragg scatteredintensity vs. the expansion time in three dimensions for three different initial latticedepths: 5ER (circles), 8ER (triangles), and 15ER (squares). Each data set is normal-ized to the Bragg intensity at the initial time. Revivals at lower lattice depths indicatecoherence of the atoms across multiple lattice sites. The lines are phenomenologicalfits to exponentially decaying sine waves. Different lattice depth data are offset forclarity and the error bar is represenative. Inset is a solution to the one-dimensionalGross-Pitaevskii equation with experimental parameters and an initial state of a chainof gaussian wavefunctions, showing the revivals of the density distribution as a func-tion of time. The white dotted line represents the revival time.
63
4 6 8 10 12 14 16
0
50
100
150
200
250
Rev
ival
Inte
nsity
(ar
b.)
4 6 8 10 12 14 16
0
0.2
0.4
0.6
0.8
1
Lattice Depth (ER
)
Nor
m. R
eviv
al In
tens
ity (
arb.
)
Figure 3-9: Bragg scattering revival as a probe of superfluid coherence. Top: Braggscattered revival signal at 60 µs vs. lattice depth. Bottom: The top plot normal-ized to the Bragg scattered signal without expansion. We see a decrease in revival,and consequently superfluid coherence, as the lattice is increased. Error bars arestatistical.
ligible effect on the decay of revivals. Empirically, randomizing the phase between
neighboring sites reduces the revival fraction. The loss of phase coherence across
lattice sites could be due to factors such as temperature, beyond Gross-Pitaevskii
equation effects, or technical factors. We have considered quantum depletion [118],
but the calculated depletion fraction at 5ER is 2%, too low to account for the observed
decay. Future work should provide a more complete picture of the decay of revivals.
Note that Bragg scattered revivals are complementary to the observation of diffrac-
tion peaks in time-of-flight absorption images [42]. Both are based on matter-wave
interference due to superfluid coherence: revivals at short expansion times and diffrac-
64
tion at long expansion times. Diffraction is a far-field effect only possible for finite
size samples, whereas revivals are a near-field bulk effect. The order of the revival
or the angular resolution of the diffraction peaks determine whether these techniques
probe short-range or long-range spatial coherence. Further studies are needed to de-
termine which technique is more sensitive to certain aspects of the superfluid to Mott
insulator transition.
In this work we have not focused on the effects of occupation number on Bragg
scattering. In principle the atomic wavefunctions are more extended for higher occu-
pation numbers, but this effect is small for our parameters. However, light scattering
at higher occupation numbers will have an inelastic component due to colliding atoms
or photoassociation [34]. The dependence of Bragg scattering on lattice depth sug-
gests that these effects are not dominant.
To examine this, we measured the Bragg reflection by varying the lattice ramp
up time, where the ramp up was up to 15ER. The Bragg reflection was essentially
unaffected by the ramp up time down to 1 ms, where we do not necessarily expect the
lattice ramp up to be adiabatic and therefore could expect the occupation number of
the atoms to be randomly distributed. This shows that the Bragg reflection is simply
a manifestation of the underlying crystal ordering, and does not depend strongly on
the exact filling of the atoms in the optical lattice.
3.6 Conclusions
After Bragg scattering has been established as a probe for Mott insulators, it can
now be applied to study other types of quantum phases such as antiferromagnetic
ordering in both the occupation number and spin sectors [21, 93]. Although tem-
peratures required to realize spin ordering are on the order of tens to hundreds of
picokelvins, recent experimental results suggest a way forward to lower the tempera-
ture of a two-component Mott insulator [76].
To detect spin ordering such as the antiferromagnetic state, we need spin-dependent
Bragg scattering. Since spins in an ultracold atom context will be different hyperfine
65
states, by choosing an appropriate wavelength of the Bragg beam one can scatter
selectively from a chosen state as well as scatter in a spin independent way [21]. Thus
Bragg scattering will allow experimentalists to probe both density and spin ordering
by varying the Bragg beam wavelength.
In conclusion, we have observed Bragg scattering of near-resonant photons from a
quantum degenerate Bose gas in a three-dimensional optical lattice. We have shown
that this technique probes not only the periodic structure of the atoms, but also the
spatial and momentum width of the localized atomic wavefunctions and the superfluid
coherence.
66
Chapter 4
Raman-Assisted Tunneling and the
Harper Hamiltonian
Ultracold atoms in optical lattices have ushered in an era of unprecedented con-
trollability, tunability, and flexibility in the study of strongly-correlated many body
physics, both experimentally and theoretically [13, 70, 42]. Here we propose and ex-
perimentally study a scheme where one can experimentally realize the Harper Hamil-
tonian with ultracold atoms in optical lattices. The Harper Hamiltonian was de-
scribed in section 2.5, and is expected to give rise to a fractal band structure called
the Hofstadter butterfly [52], shown in Fig. 2-7, mimicing solids in extremely high
magnetic fields of order one billion gauss. Once interactions are included, the Harper
Hamiltonian is believed to give rise to fractional quantum Hall states, which would
allow the experimental study of fascinating topological phases of matter [94, 48].
We have emphasized how ultracold atomic systems can simulate the behavior of
electrons in solid state materials. However many interesting effects, such as the inte-
ger and fractional quantum Hall effects [107, 68], occur when a solid is placed in an
external magnetic field. Due to the charge neutrality of ultracold atoms, one cannot
directly apply a magnetic field to a cold atom system and expect a similar response
as an electron in a solid. Despite this seeming limitation, there are ways to create
effective magnetic fields which lead to Hamiltonians as if the atoms were moving in
a magnetic field with an effective charge. The method initially realized by experi-
67
mentalists was to physically rotate the atomic cloud to genereate rotational velocity,
which then acted as an effective magnetic field giving rise to quantized vortices in bulk
Bose-Einstein condensates [74, 1]. The seminal work by Ian Spielman’s group [72]
realized effective magnetic fields by taking advantage of Raman transitions between
different hyperfine states, thus paving the way for optical realizations of effective
magnetic fields.
Other methods to generate artificial gauge fields have been proposed by Jaksch and
Zoller [59] and Gerbier and Dalibard [37], which involve Raman transitions but now
in an optical lattice geometry. Both schemes envision a two-component system, where
the lattices of the two components are offset by half a lattice spacing. The Raman
transitions induce tunneling between the two sub-lattices, in the process imprinting
a position-dependent phase.
Furthermore, there have been experimental realizations of a spatially-independent
phase by the Sengstock [101] and Spielman groups [60], as well as a staggered flux
arrangement by the Bloch group [3]. The Sengstock group realizes their phase by a
particular type of lattice phase modulation, the Spielman group uses Raman transi-
tions and radiofrequency magnetic fields, and the Bloch group uses a superlattice in
conjunction with Raman beams. Although all three results have led to new insights
into the nature of phases in ultracold atomic systems, they have limitations if one is
hoping to reach high enclosed fluxes per unit cell. In particular, the Sengstock and
Spielman approaches as it is only create a spatially-independent phase, which does
not give rise to any enclosed flux. In the Bloch approach, because the superlattice
can only create a staggered flux arrangement, no globally macroscopic enclosed flux
is possible. In all three of these cases, rectification of the fluxes to create a spatially
varying flux seem to be in principle possible, but the proposed schemes involve many
more laser beams and requires further complexity in the experimental arrangement.
In contrast, our proposed technique involves only two lattice laser beams and two
running wave Raman beams and a single spin state of bosonic 87Rb, which is signifi-
cantly more simple than the other proposed schemes and should allow access to any
desired enclosed flux with simple modifications to the experimental geometry.
68
Figure 4-1: Schematic of a tilted lattice and Raman beams to create artificial gaugefields. The tilt induces an energy offset ∆ between neighborin sites. Then Ramanbeams that are frequency detuned by exactly ∆ can cause transitions between local-ized states in nearest neighbors. Furthermore, this tunneling can have a position-dependent phase which gives rise to topologically non-trivial ground states.
Our proposal involves realizing the topologically non-trivial Harper Hamiltonian
by creating a two-dimensional optical lattice with a linear potential gradient along
one of the lattice directions. As it is, this tilted potential causes a suppression of
resonant tunneling along the tilt direction and only allows resonant tunneling along
the direction perpendicular to the tilt. Then on top of that two laser beams are
applied which are frequency detuned to precisely match the energy offset caused by
the linear potential. Then these two laser beams are able to induce a Raman transition
between a localized state in one lattice site to a nearest neighbor lattice site in the tilt
direction, effectively flattening out the tilt. Thus we call these two beams our Raman
beams. Crucially, in addition to restoring tunneling, these running wave Raman
beams can imprint a position-dependent phase on the atoms, which then results in
the Harper Hamiltonian. A schematic of this setup is shown in Fig. 4-1. Note that
the tilt can be created by any physical mechanism which creates a linear potential
gradient, including gravity and magnetic field gradients.
69
In the rest of this chapter, I will describe in more detail the theoretical underpin-
nings of the scheme, as well as present experimental results which give indications
that we have successfully created the Harper Hamiltonian. However, thus far we have
been unable to observe direct signatures of the ground state of the Harper Hamilto-
nian, namely superfluid interference peaks corresponding to the band structure of the
Harper Hamiltonian.
4.1 Raman-Assisted Tunneling in the Perturba-
tive Limit
Modification of tunneling in an optical lattice system by dynamical modulation
of the lattice have been explored by other groups, notably the Arimondo and Tino
groups in Italy [71, 92, 56, 6, 5, 103]. Our proposed Raman beams are in the same
spirit, since two laser beams which induce a Raman transition can be thought of as
a travelling wave optical lattice.
In particular, two Raman lasers with wavevectors k1 and k2 and lattice depth VK
can induce tunneling through a time-dependent potential
VK(r, t) = VK cos2(q · r− ωt/2) (4.1)
= VK1 + cos(2q · r− ωt)
2(4.2)
=VK4
(1 + ei(2q·r−ωt) + e−i(2q·r−ωt)
)(4.3)
=VK4
(1 + ei(δk·r−ωt) + e−i(δk·r−ωt)
), (4.4)
where ω is the frequency detuning between the two Raman beams, 2q = δk, δk =
k1 − k2, and q = (k1 − k2)/2.
Let us assume that the linear tilt exists in the x direction with an energy offset
between nearest neighbor lattice sites of ∆ and that the detuning of the Raman beams
is exactly ∆. Then in the perturbative limit of VK ∆ and in the rotating wave
70
approximation, the laser-assisted tunneling term can be written
K =VK4
∫d2rw∗(r−R)w(r−R− dx)e
−iδk·r (4.5)
=VK4
∫d2rw∗(r)w(r− dx)e
−iδk·(r+R) (4.6)
= e−iδk·RVK4
∫d2rw∗(r)w(r− dx)e
−iδk·r (4.7)
= e−iδk·RVK4
∫dxw∗(x)w(x− dx)e−iδkxx (4.8)
×∫dxw∗(y)w(y)e−iδkyy (4.9)
= K0e−δk·R, (4.10)
where K0 = VK/4∫d2rw∗(r)w(r−dx)e
−iδk·r, and w(r) = w(x)w(y) where w(x) is the
Wannier-Stark state due to the linear tilt in the x direction and w(y) is the Wannier
function. The position-dependent phase which gives rise to the Harper Hamiltonian is
embodied in the factor δk·R. I plot the absolute value of the relevant overlap integrals
present in the Raman-assisted tunneling expression in Fig. 4-2. The overlap integrals
teach us something interesting about the requirements to achieve position-dependent
phases under realistic conditions.
We see that the overlap integral involving the Wannier-Stark states in the tilt
direction exhibits oscillatory behavior. In particular, we learn that there must be a
non-zero momentum transfer in the tilt direction for the Raman-assisted tunneling to
be strong enough to induce tunneling. This is an important point, since from phase
accumulation considerations, all one needs is momentum transfer in the direction
orthogonal to the tilt. Thus to get position-dependent phase which gives rise to non-
zero enclosed flux and a sizable tunneling amplitude, the Raman beams must transfer
momentum along both the direction of the tilt and orthogonal to it.
Note that VK can be related to the two-photon Rabi frequency of the two Ra-
man lasers, which is also related to the one-photon Rabi frequency of the individual
lasers. Knowing the precise relationship between the lattice depth and two-photon
Rabi frequency is important when we calibrate the strength of our Raman beams by
71
0 2 4 6 8 100
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
k (2π/λ)
Ove
rlap
Inte
gral
⟨0|eikx|1⟩
0 2 4 6 8 100
0.2
0.4
0.6
0.8
1
1.2
1.4
k (2π/λ)
Ove
rlap
Inte
gral
⟨0|eiky|0⟩
Figure 4-2: Overlap integrals for perturbative Raman coupling. The integral in thex direction is done with Wannier-Stark states and the integral in the y direction isdone with Wannier states.
measuring the two-photon Rabi oscillation frequency between states with momentum
0 and k1 − k2.
The two-photon Rabi frequency ΩR2 is given as [59]
ΩR2 =ΩRgΩRe
2δ(4.11)
where ΩRg,e are the one-photon Rabi frequency of each of the Raman beams and δ is
the detuning between the Raman laser and the D1 and/or D2 lines. The one-photon
Rabi frequency is given by ~Ω = eE〈e|z|g〉 where e is the electron charge, E is the
electric field, and e〈e|z|g〉 ≡ D is the electric dipole moment [35]. The two-photon
Rabi oscillation from the ground state |g〉 to an excited state |e〉, starting in |g〉 is
given by Pe(t) = sin2(ΩR2t/2). This means that the population transfer oscillates at
frequency ΩR2.
Now, the moving lattice depth VK is given by VK = ~ΩRgΩRe/δ [80]. This can
be understood by recalling that the single traveling wave AC Stark shift is given as
~Ω2Rg/(4δ). Interference of two laser beams leads to a multiplicative factor of 4. So,
combining all that, the two-photon Rabi oscillation frequency ΩR2 is related to the
72
Raman lattice depth VK as
VK = 2ΩR2. (4.12)
Note that in general K0 is complex and so can have some non-zero phase in
addition to the spatially-dependent phase give by δk·R. But for a given configuration,
the phase is not spatially dependent so is effectively a global, constant phase.
4.1.1 Energy Hierarchy to Realize the Harper Hamiltonian
Certain inequalities must be satisfied in order to realize the Harper Hamiltonian.
First, the band gap ω must be larger than any other energy scale so that the atoms
do not occupy higher orbital states. In particular, if the potential tilt ∆ is equal
to the band gap, Wannier-Stark transitions, where states in the lowest band in one
lattice site become resonant with states in the first excited band in a neighboring site,
will be allowed to happen, therefore destroying the requirement to be in the lowest
band [45, 40]. Therefore the potential tilt ∆ must be smaller than ω.
A second requirement is for the bare tunneling energy J to be smaller than the
potential tilt ∆. If this condition is not satisfied, then resonant tunneling can be
stronger than the light-assisted tunneling and consequently position-dependent phase
upon tunneling will not manifest itself.
These two conditions lead to the following energy hierarchy,
J ∆ ω. (4.13)
For 87Rb, the mass is 1.443 × 10−25 kg [97], and for a lattice spaced apart 532
nm, the potential tilt due to gravity is 1137 Hz. Similar tilts can also be achieved
for atoms in the hyperfine state |F = 2,mF = −2〉 with a magnetic field gradient of
around 15 G/cm, which is experimentally accessible. For lattice depths of 10ER, the
tunneling energy J is around 45 Hz and the band gap is around 10 kHz. Thus these
experimentally accessible parameters will satisfy the inequalities necessary to realize
73
the Harper Hamiltonian.
4.2 Experimental Geometries to Realize Specific α
As shown in section 4.1, the factor δk · R gives rise to the non-trivial topologi-
cal phase in the Harper Hamiltonian. Therefore, controlling the angle between the
Raman beams and the two-dimensional lattice axes are crucial to realize the Harper
Hamiltonian with a particular α. First I explore the possible α for an arbitrary Ra-
man geometry in a square lattice, and then focus on a specific geometry which we
have implemented which should give rise to α = 1/2.
Let us consider a 2D geometry where the lattice and Raman beams are propagating
in a 2D plane as in Fig. 4-3. Let the lattice be square along the x and y axes and
have spacing a, so that the position vector is R = a(mx+ny). Also, let the difference
in k vectors of the two Raman beams k1 and k2 be k = k1 − k2 so that if the angle
between k1 and k2 is θ and k1 = k2 = 2π/λ, then |k| ≡ k = 4π/λ| sin(θ/2)|. Also, in
this lattice coordinates write k = k(cosφx+ sinφy). The phase is
k ·R = 2π
∣∣∣∣sin(θ2)∣∣∣∣ (m cosφ+ n sinφ) (4.14)
= mφx + nφy, (4.15)
where we defined φx = 2π| sin(θ/2)| cosφ and φy = 2π| sin(θ/2)| sinφ. Now assume
the tilt is in the x direction. Then the phase accumulated is as given in the left of
Fig. 4-3. We see that the flux accumulated along one unit cell is Φ = φy, which
corresponds to a magnetic flux per plaquette of α = φy/2π. If the tilt is in the y
direction, then the phase accumulated is as given in the right of Fig. 4-3. We see that
the flux accumulated along one unit cell is Φ = φx which corresponds to a magnetic
flux per plaquette of α = φx/2π. Thus we see that if the linear tilt is along a given
direction, the total phase accumulated after one full rotation around a plaquette is
given by the phase that depends on the perpendicular direction.
74
m
n
3
2
1
21 3 x
y
Raman
Beams
yx ! yx !2
yx 2 yx !! 22
yx !! 3 yx !! 32
!2k
1k
Linear
Tilt
21kkk
!
(a) k
n
3
2
1
21 3
yx ! yx !2
yx 2 yx !! 22
yx !! 3
yx 23
Linear
Tilt
m x
y
(b)
Figure 4-3: Tilt and phase accumulated for general Raman geometry. Left figure istilt along the x direction, the right figure is tilt along the y direction. Note φx =2π| sin(θ/2)| cosφ and φy = 2π| sin(θ/2)| sinφ.
4.2.1 Experimental Geometry for α = 1/2
Now consider the specific case where the Raman beams are copropagating along
the lattice beam directions so that k1 = 2π/λx and k2 = 2π/λy. A schematic is
shown in Fig. 4-4. In this case, we have θ = 90 and φ = −45, so we get that
k ·R = π(m− n). (4.16)
With a linear tilt along the x direction, the total phase accumulated around one
plaquette is given by Φ = π. This phase then corresponds to a flux density per
plaquette of α = 1/2 and will be the geometry I will be focusing on for most of the
rest of this chapter.
4.3 Raman-Assisted Tunneling in the Rotating Frame
In section 4.1 we showed the effective Raman-assisted tunneling and accompanying
position-dependent phase factor which gives rise to effective magnetic fields in the
perturbative regime. Here I derive a more general expression for the Raman-assisted
tunneling which include both the position-dependent phase and saturation of the
tunneling amplitude. The effective tunneling can be solved for more general cases by
75
m
n
3
2
1
2 1 3
2k
21kkk
!
1k
0
0
0
x
y
Linear
Tilt
Figure 4-4: Experimental geometry for α = 1/2. The phase accumulated along thetilt alternates between 0 and π.
rewriting the original Hamiltonian into a time-independent Hamiltonian with effective
tunneling by using the rotating-wave approxiation [103].
If you have a two-dimensional system with a square 2D lattice VL(x, y) with lattice
spacing a and a linear tilt with energy per lattice site of ∆ in the x direction and an
additional term E(x, y, t), we can write the Hamiltonian as
H =p2x
2m+
p2y
2m+ VL(x, y) +
∆
ax+ E(x, y, t). (4.17)
This can be rewritten into a second quantized, time-dependent Hamiltonian of the
form [103]
H =∑m,n
∆m|m,n〉〈m,n| − Jy∑m,n
(|m,n+ 1〉〈m,n|+ c.c.)
+∑
m,m′,n,n′
|m,n〉〈m,n|E(x, t)|m′, n′〉〈m′, n′|, (4.18)
76
−2 −1 0 1 2 3−1
−0.5
0
0.5
1
1.5
2
2.5
Position (λ/2)
Wav
efun
ctio
n an
d P
oten
tial (
arb.
)
Latticecoskxsinkx⟨x|0⟩⟨x|1⟩
Figure 4-5: Tilt and Wannier-Stark states. Sines and cosines are also depicted to geta better sense for how the overlap integrals should behave.
where |m,n〉 ≡ |m〉x ⊗ |n〉y, |m〉x is the Wannier-Stark state centered at site m
in the x direction [45, 40], |n〉y is the Wannier state centered at site n in the y
direction, and Jy is the bare tunneling in the y direction just as in the Bose-Hubbard
Hamiltonian. The Wannier-Stark states are shown in Fig. 4-5. From this Hamiltonian,
a unitary transformation can be performed which eliminates the time-dependence in
the rotating wave approximation.
The following identities are necessary to proceed and can be derived from the
definition of exponentiated operators and the properties of bra and ket operators.
〈m|n〉 = δmn (4.19)
(|m〉〈m|)n = |m〉〈m| (4.20)
e−ix|m〉〈m|eix|m〉〈m| = |m〉〈m| (4.21)
e−i∑m x|m〉〈m| =
∑m
e−ix|m〉〈m| =∑m
e−ix|m〉〈m|. (4.22)
Our Raman beams create a potential of the form
E(x, y, t) = Ω sin(2q · r− ωt). (4.23)
Recall that 2q = k1 − k2 and for the specific case under consdieration k1 = kx,
77
k2 = ky and k = 2π/λ. The relevant matrix elements are then 〈m,n| sin(2q · r −ωt)|m + l, n + p〉. With a change of variables using the symmetry of the system, we
can write
〈m,n| sin(2q · r− ωt)|m+ l, n+ p〉 = 〈0, 0| sin(2q · (r + Rm,n)− ωt)|l, p〉, (4.24)
where Rm,n = max + nay. In general, we can write 2q · r = kxx + kyy. We let
θm,n = ωt− 2q ·Rm,n and φm,n = 2q ·Rm,n. Then we get the matrix element in the
x direction to be
〈0, 0| sin(2q · (r + Rm,n)− ωt)|l, 0〉 = 〈0, 0| sin(kxx+ kyy − θm,n)|l, 0〉 (4.25)
= 〈0, 0| sin kxx cos kyy cos θm,n (4.26)
+ sin kxx sin kyy sin θm,n (4.27)
+ cos kxx sin kyy cos θm,n (4.28)
− cos kxx cos kyy sin θm,n|l, 0〉 (4.29)
= − sin θm,n〈0| cos kxx|l〉. (4.30)
Note that we have used the approximations 〈0| sin kxx|l〉 = 〈0| sin kyy|0〉 = 0, due to
the asymmetry in the Wannier-Stark state and 〈0| cos kyy|0〉 = 1. We are primarily
interested in nearest neighbor tunneling, so we need only consider l = 0, 1. Then
we can approximate 〈0| cos kxx|0〉 = 1 and 〈0| cos kxx|1〉 = −2Jx/∆. A numerical
evaluation of these integrals shown in Fig. 4-6 support these approximations.
Similar analysis shows that the matrix element in the y direction becomes
〈0, 0| sin(2q · (r + Rm,n)− ωt)|0, 1〉 = cos θm,n〈0| sin kyy|1〉, (4.31)
where we used the approximations 〈0| sin kxx|0〉 = 〈0| cos kyy|1〉 = 0, and 〈0| cos kxx|0〉= 1. Now numerically for typical experimental parameters of lattice depth 10ER and
tilt per lattice site of 1 kHz, we find that 〈0| sin kyy|1〉 ≈ 0.01. The off-diagonal term
in the y direction then will have a component with strength Jy and 0.01Ω cos θm,n.
78
0 2 4 6 8 100
0.2
0.4
0.6
0.8
1
1.2
1.4
k (2π/λ)
Ove
rlap
Inte
gral
⟨0|coskx|0⟩⟨0|cosky|0⟩
0 2 4 6 8 100
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
k (2π/λ)
Ove
rlap
Inte
gral
⟨0|sinkx|0⟩⟨0|coskx|1⟩⟨0|sinkx|1⟩⟨0|sinky|0⟩⟨0|cosky|1⟩⟨0|sinky|1⟩2J/∆
Figure 4-6: Overlap integrals to calculate Raman-assisted tunneling. The integralsinvolving x are done with Wannier-Stark states and the integrals involving y are donewith Wannier states. The calculations were done assuming a lattice depth of 10ER
in both x and y directions the linear tilt along x was assumed to be 1 kHz per latticesite.
To be able to neglect this time-dependent term, we need Jy 0.01Ω. For lattice
depth 10ER, we have Jy = 45 Hz and typically Ω ≈ 1 kHz, so we marginally have
Jy 0.01Ω. In any case, in the perturbative regime of Ω/∆ 1, this time-dependent
term will be even less important.
So the relevant terms in the Hamiltonian become
〈m,n| sin(2q · r− ωt)|m,n〉 = − sin(ωt− 2q ·Rm,n) (4.32)
〈m,n| sin(2q · r− ωt)|m+ 1, n〉 =2Jx∆
sin(ωt− 2q ·Rm,n) (4.33)
〈m,n| sin(2q · r− ωt)|m,n+ 1〉 = 0. (4.34)
So our Hamiltonian becomes
H =∑m,n
(∆m−Ω sin θm,n)|m,n〉〈m,n|+∑m,n
[2ΩJx
∆sin θm,n|m+ 1, n〉〈m,n|+ h.c.
]−∑m,n
[Jy|m,n+ 1〉〈m,n|+ h.c.] . (4.35)
79
To get rid of our time-dependence, we can define a unitary transformation
U = exp
[−i∑m,n
mωt+
Ω
~ωcos θm,n
|m,n〉〈m,n|
](4.36)
= exp
[−i∑m,n
Λm,n|m,n〉〈m,n|]
(4.37)
=∑m,n
exp [−iΛm,n] |m,n〉〈m,n| (4.38)
where Λm,n = mωt+ Ω~ω cos θm,n. Then in the rotating frame, the Hamiltonian trans-
forms as
H ′ = UHU ′ − i~U dUdt. (4.39)
The time derivative term in the transformed Hamiltonian gets rid of the time depen-
dence in the diagonal term and replaces ∆ with δ ∼ ∆−ω, which becomes zero if you
drive the system on resonance, which is the situation we are considering. Since U is
diagonal, it has no effect on the diagonal terms of H. So we can focus our attention
on the off-diagonal terms of the transformation.
First we consider how the off-diagonal coupling in the x direction transform.
UHxoff−diagU
′ =
∑m′,n′
exp [−iΛm′,n′ ] |m′, n′〉〈m′, n′|
2ΩJx∆
sin θm,n|m+ 1, n〉〈m,n|
(4.40)
×∑m′′,n′′
exp [iΛm′′,n′′ ] |m′′, n′′〉〈m′′, n′′|
(4.41)
= exp [−iΛm+1,n] |m+ 1, n〉2ΩJx∆
sin θm,n exp [iΛm,n] 〈m,n| (4.42)
=2ΩJx
∆sin θm,ne
i(Λm,n−Λm+1,n)|m+ 1, n〉〈m,n|. (4.43)
Keep in mind the complex conjugate term exists but behaves almost identically. The
80
exponential term can be simplified as
ei(Λm,n−Λm+1,n) = exp i
[mωt+
Ω
~ωcos θm,n − (m+ 1)ωt− Ω
~ωcos θm+1,n
](4.44)
= exp i
[−ωt+
Ω
~ω(cos θm,n − cos θm+1,n)
](4.45)
= exp i
[−ωt− 2
Ω
~ωsin
(kxa
2
)sin
(θm,n −
kxa
2
)](4.46)
= e−iωteiΓ sin(θm,n−kxa/2) (4.47)
= e−iωt∑p
Jp(Γ)eip(θm,n−kxa/2). (4.48)
where Γ = −2 Ω~ω sin
(kxa2
)and with the relationship eiΓ sin θ =
∑p Jp(Γ)eipθ. With
this, the off-diagonal term in the x direction can be written
2ΩJx∆
sinφm,nei(Λm,n−Λm+1,n) =
2ΩJx∆
eiωt−iφm,n − e−iωt+iφm,n2i
(4.49)
× e−iωt∑p
Jp(Γ)eip(ωt−φm,n−kxa/2) (4.50)
=ΩJxi∆
(e−iφm,nJ0(Γ)− e−iφm,n−ikxaJ2(Γ)
)(4.51)
=ΩJxi∆
e−iφm,n (J0(Γ) + J2(Γ)) (4.52)
=ΩJxi∆
e−iφm,n2
ΓJ1(Γ) (4.53)
= −i~ωJx∆
e−iφm,nJ1
(2Ω
~ω
)(4.54)
where we used kxa = π, the identity J0(x) + J2(x) = 2/xJ1(x), and Γ = −2 Ω~ω and
neglected all terms which have time dependence.
Now let us consider the off-diagonal term corresponding to tunneling along the y
81
direction which is the direction perpendicular to the tilt.
UHyoff−diagU
′ = −∑m′,n′
exp [−iΛm′,n′ ] |m′, n′〉〈m′, n′|Jy|m,n+ 1〉〈m,n| (4.55)
×∑m′′,n′′
exp [iΛm′′,n′′ ] |m′′, n′′〉〈m′′, n′′|
(4.56)
= − exp [−iΛm,n+1] |m,n+ 1〉Jy exp [iΛm,n] 〈m,n| (4.57)
= −Jyei(Λm,n−Λm,n+1)|m,n+ 1〉〈m,n|. (4.58)
The exponential term in this case becomes
ei(Λm,n−Λm,n+1) = exp i
[mωt+
Ω
~ωcos θm,n −mωt−
Ω
~ωcos θm,n+1
](4.59)
= exp i
[Ω
~ω(cos θm,n − cos θm,n+1)
](4.60)
= exp i
[−2
Ω
~ωsin
(kya
2
)sin
(θm,n −
kya
2
)](4.61)
= eiΓ sin(θm,n−kya/2) (4.62)
=∑p
Jp(Γ)eip(θm,n−kya/2). (4.63)
where we used the relationship eiΓ sin θ =∑
p Jp(Γ)eipθ and Γ = −2 Ω~ω sin
(kya
2
).
Then keeping only time independent terms, we have
−Jyei(Λm,n−Λm,n+1) = −Jy∑p
Jp(Γ)eip(θm,n−kya/2) (4.64)
= −JyJ0(Γ) (4.65)
= −JyJ0
(−2
Ω
~ωsin
(kya
2
))(4.66)
= −JyJ0
(2Ω
~ω
), (4.67)
where we used the fact kya = π and J0(x) = J0(−x).
82
In the case of resonant driving, ~ω = ∆ so ultimately we obtain
H ′ =∑m,n
[−ie−iφm,nJxJ1
(VK∆
)|m+ 1, n〉〈m,n|+ h.c.
]−∑m,n
[JyJ0
(VK∆
)|m,n+ 1〉〈m,n|+ h.c.
]. (4.68)
4.3.1 Perturbative and Exact Raman-Assisted Tunneling
So far we have considered two approaches to determining the laser-assisted tun-
neling. One was perturbative, giving rise to a Raman-assisted tunneling term which
depended linearly on the Raman lattice depth, and the second involved a more explicit
treatment of the time-dependent term which showed saturation in the magnitude.
Both approaches give rise to the position-dependent phase which is crucial in
realizing artificial gauge fields. However, the dependence of the magnitude of Raman-
assisted tunneling on the Raman lattice depth is different for the two approaches. The
two expressions are
Kpert =VK4
∫dxw∗(x)w(x− dx)e−iδkxx (4.69)
×∫dxw∗(y)w(y)e−iδkyy (4.70)
|Kexact| = J
∣∣∣∣J1
(VK∆
)∣∣∣∣ . (4.71)
We can numerically check how close the two magnitudes are. This is shown in Fig. 4-
7, where we had a lattice depth of 10ER and an energy tilt between adjacent sites of
1 kHz. We see good agreement between the two results for VK/∆ . 2.
4.4 Amplitude and Phase Modulation Tunneling
It is possible to relate the mechanism of Raman-assisted tunneling to amplitude
and phase modulation of the underlying lattice.
83
0 0.5 1 1.5 2 2.5 30
10
20
30
40
50
VK/∆
|K| (
Hz)
Exact KPerturbative K
Figure 4-7: Perturbative and exact treatment of Raman-assisted tunneling. We as-sumed a lattice depth of 10ER and an energy tilt between adjacent sites of 1 kHz.
84
In the case of amplitude modulation, in 1D the Hamiltonian can be written [103]
HAM =p2x
2m+ VL(x)(1 + α sinωt) +
∆
ax, (4.72)
where α is the modulation fraction and is such that 0 ≤ α ≤ 1 and we write VL(x) =
−U0/2 cos(2kx) where k = 2π/a. In second quantized form in the Wannier-Stark
basis this Hamiltonian becomes
HAM = −∑
∆n|n〉〈n|+∑n
[αU0
2〈n+ 1| cos 2kx|n〉 sinωt|n+ 1〉〈n|+ h.c.
].
(4.73)
Then a unitary transformation similar to what I performed in the previous section
gives, on resonance at ~ω = ∆,
H ′AM =∑n
[iαU0
4〈n+ 1| cos 2kx|n〉|n+ 1〉〈n|+ h.c.
]. (4.74)
Thus we see that the bare Hamiltonian in the Wannier-Stark basis has time-dependent
off-diagonal terms but time-independent diagonal terms, which in the rotating wave
can be written as a Hamiltonian with effective time-independent tunneling. In partic-
ular, the result permits a unlimited increase in the effective tunneling amplitude if one
drives the modulation harder and harder in the frame where the lattice is stationary.
In the case of phase modulation, the lattice in the laboratory frame can be written
as V = −U0/2 cos[2k(x − x0 sinωt)] where x0 and ω are the modulation amplitude
and frequency [75]. In a frame accelerating at exactly the phase modulation, the
Hamiltonian can be written [71, 92, 56, 103]
HPM =p2x
2m+ VL(x) + βx sinωt+
∆
ax, (4.75)
where we have the atomic mass m and β = mx0ω2. With this, in the Wannier-Stark
85
basis the Hamiltonian becomes
HPM = −∑
n(∆ + βa sinωt)|n〉〈n| −∑n
[β〈n+ 1|x|n〉 sinωt|n+ 1〉〈n|+ h.c.] .
(4.76)
Then in the appropriate rotating frame the Hamiltonian becomes
H ′PM =∑n
[i∆〈n+ 1|x|n〉
aJ1
(βa
∆
)|n+ 1〉〈n|+ h.c.
]. (4.77)
Note that in this case the effective tunneling amplitude has a maximum possible value
limited by the maximum of the Bessel function as in Raman-assisted tunneling [92].
4.5 Experimental Studies of Amplitude Modula-
tion (AM) Tunneling
Due to the fundamentally related phenomenon of ampltidue modulation (AM)
tunneling and Raman-assisted tunneling, we have used amplitude modulation as a
testing ground to understand under what parameter regimes we observe Raman-
assisted tunneling. Experimentally, AM is much easier to implement since it can
be implemented in 1D and does not require any other laser beams aside from that
which is creating the lattice. AM tunneling has been experimentally explored by
other groups, most notably the group of Guglielmo Tino [6, 5, 103].
4.5.1 Experimental Sequence for AM Tunneling
For our initial experiments, we have used gravity as our linear potential tilt, which
should create an energy offset per lattice site with 532 nm spacing of 1137 Hz. Our
experimental sequence is as follows. Using standard optical and evaporative cooling
techniques [100] to achieve a Bose-Einstein condensate of a few hundred thousand
atoms, we transfer the atoms which are in the |F = 1,mF = −1〉 hyperfine state into
a crossed optical dipole trap with trap frequencies in the range of 30 to 160 Hz. At
86
that point, we adiabatically ramp up a 1D optical lattice in the direction of gravity
which is created by retroreflecting a single laser beam in about 200 ms. Then we
extinguished our crossed dipole traps in 200 ms (we also attempted to extinguish
even faster, but aside from center-of-mass motion perpendicular to the lattice, this
had no discernable effect on the results presented here). At this point, the atoms
are held only by an optical lattice pointing in the vertical direction. Confinement
in the vertical direction is provided by the optical lattice, and confinement in the
perpendicular direction is provided by the gaussian envelope of the lattice beam.
Furthermore, by extinguishing the crossed optical dipole traps, that automatically
creates a linear gradient with exactly the strength of gravity. The lowering of the
confinement in the perpendicular direction appears to help to reduce density and
inhomogeneous tilts due to curvature. At this point, we begin to modulate the lattice
depth and perform measurements.
4.5.2 In Situ Measurements of AM Tunneling
Our initial measurements were performed in situ, by taking absorption images of
the atoms in the optical lattice. When the potential gradient is present, the atoms are
unable to tunnel because they are off-resonant. Modulation of the lattice amplitude
reestablishes tunneling as described in a previous section. This AM tunneling then
allows the atom cloud to expand symmetrically in the vertical direction due to the
lack of confinement in addition to the lattice. Thus measuring the spatial width of
the atomic cloud after modulation gives an indication of the strength of tunneling.
The way we determined the resonance frequency of AM tunneling was to modulate
an originally static lattice with a depth of 10ER for 501 cycles of modulation with
an amplitude of 2ER and repeat the experiment for various modulation frequencies.
This result is shown in the top left of Fig. 4-8. Fitting the data to a lorentzian,
we determine the resonance to be at 1133.5 ± 0.5 Hz and the width to be 2.2 ± 0.2
where the uncertainties indicate 95% confidence intervals (note this is equivalent to
two standard deviations). From the fact that g = 9.81 m/s2, we expect the resonance
at 1137 Hz. The discrepancy is primarily expected to be due to slight misalignment
87
9 10 11 12 13 14 15 16 175
10
15
20
25
Lattice Depth (ER)
In S
itu W
idth
(µm
)
0 200 400 600 800 10000
10
20
30
40
50
60
Modulation Cycles at 1133.5 Hz
In S
itu W
idth
(µm
)
0 1 2 3 4 5 6 70
5
10
15
20
Modulation Amplitude (ER)
In S
itu W
idth
(µm
)
Figure 4-8: In situ amplitude modulation tunneling measurements. Top left: In situwidth of the atomic cloud as a function of AM frequency. Inset is a representativeatom distribution on and off resonance observed by absorption imaging. Top right:Atom width as a function of lattice depth for modulation on resonance at 1133.5 Hz.Curve is a plot of the square of the analytical expression for the tunneling matrixelement, whose close agreement with data suggests that the width increases as thesquare of the tunneling matrix element. Bottom left: Atom width as a function of thenumber of modulation cycles. The curve is a fit of the data to a parabola. Bottomright: Atom width as a function of modulation amplitude. The curve is a fit of thedata to a parabola. Both bottom curves suggest the width increases as the square ofthe tunneling matrix element, (aJt)2.
88
of the laser beams.
We also studied the dependence of the in situ width on the lattice depth. Our
static lattice was at a depth of 10ER, on resonance at 1133.5 Hz and modulated for
501 cycles at an amplitude of 2ER. This is shown in the top right of Fig. 4-8. I
also plot the analytic expression for the square of resonant tunneling with only one
fitting parameter which determines the overall proportionality factor. The analytic
expression for resonant tunneling in a flat lattice with lattice depth VL is given as [29]
J =4√πE
1/4R V
3/4L e
−2
√VLER . (4.78)
We can write VL = NLER, so in the figure I plot the square of this up to some
proportionality factor C, as
σ(VL) = CV3/2L e−4
√NL . (4.79)
It is reasonable to expect the width to increase as a function of the value aJt where
t is the modulation time and a is the lattice spacing. The close agreement between
data and the analytic expression suggests that the width increases as the square of
the tunneling, σ ∝ (aJt)2. The following two data sets are consistent with tunneling
increasing the size as (aJt)2.
We further studied the in situ width as a function of the number of modulation
cycles and modulation amplitude. These are plotted in the left bottom and right
bottom of Fig. 4-8, respectively. The data are well-fitted by parabolic curves, again
indicating that the width increases as the square of tunneling, or (aJt)2.
Note that in general, ballistic expansion goes linearly in time, and diffusive ex-
pansion goes as the square root of time. Thus the expansion behavior observed in
our measurements is not compatible with either, and could be due to heating effects.
Further study is needed to fully understand the expansion behavior of the atomic
cloud.
89
501.00 501.25 501.50 501.75 502.00 Cycles
10ER
Lattice
Depth
TOF
Images
Figure 4-9: Evolution of superfluid peaks under lattice amplitude modulation. Thesedata were taken at a lattice depth of 10ER and modulation amplitude of 2ER. Notethat the particular interference pattern depends on the particular phase of the mod-ulation when the atoms are released from the lattice.
4.5.3 Time-of-Flight Measurements of AM Tunneling
We have further studied the effects of AM tunneling on the coherence properties
of the atomic cloud by the method of time-of-flight absorption imaging. In particular,
initially after the tilt is turned on, the atomic cloud does not exhibit sharp interference
peaks, but after typically about 100 ms of amplitude modulation on resonance, we
are able to restore superfluid peaks. The speed of restoring superfluid peaks depend
on the AM strength as well as the initial lattice depth. One particular sequence of
pictures is shown in Fig. 4-9. These data were taken at a lattice depth of 10ER and
modulation amplitude of 2ER. Note that the particular interference pattern depends
on the particular phase of the modulation when the atoms are released from the
lattice, in agreement with the qualitative behavior seen by the Sengstock group in
their phase modulated phase imprinting scheme [101]. This indicates that when we
are able to establish superfluid coherence with Raman beams, the particular time-of-
flight interference pattern will depend on the exact phase relationship between the
two Raman beams, which could change from shot-to-shot.
90
4.6 Experimental Observation of Raman-Assisted
Tunneling
After having determined the experimental parameter space where we can ob-
serve amplitude modulation tunneling, we implemented the Raman-assisted tunneling
scheme to create position-dependent phases. We analyzed in situ atomic dynamics,
whose expansion we found to become saturated and subsequently decreased in the
tunneling amplitude upon increasing our Raman laser intensities, just as predicted
from our theory of Raman-assisted tunneling. Although this gives an indication that
we have successfully created the Harper Hamiltonian, we have been unable to observe
the signature of the ground state of this Hamiltonian which should manifest itself in
a time-of-flight picture.
4.6.1 Experimental Sequence for Raman-Assisted Tunneling
Our experimental sequence is identical to the case of AM tunneling described
in subsection 4.5.1 up until we create the linear potential tilt by extinguishing the
crossed optical dipole trap, except for the fact that instead of a 1D lattice, now a
2D lattice is adiabatically ramped up in the same amount of time to 10ER in both
directions (unless otherwise noted). Then the two Raman beams are turned on to a
value when combined yield an effective travelling Raman lattice depth of VK . The
experimental schematic is shown on the left of Fig. 4-10.
The Raman lattice strengths were calibrated by performing two-photon Rabi os-
cillations between momentum states of k = 0 and k = k1 − k2, where k1 = 2π/λx
k2 = 2π/λy are the wavevectors of the two Raman beams. This oscillation can be
induced by detuning the two Raman beams by 2ER/h = 4056 Hz. This can be
understood from the fact that Raman beams impart a momentum on the atoms of
magnitude |k| =√
22π/λ. The corresponding kinetic energy of the recoiling atom is
given by ~2k2/(2m) = 2ER. The oscillation frequency Ω can be related to the Raman
lattice depth as VK = 2Ω.
91
Gravity
Raman Beam
Ra
ma
n B
ea
m
a = 532 nm
1000 1100 1200 1300 14005
5.5
6
6.5
7
7.5
8
8.5
9
9.5
Raman Detuning (Hz)
In−
Tra
p W
idth
(µm
)
Figure 4-10: Raman-assisted tunneling setup and spectrum. Left: Schematic of ourexperimental setup. The energy offset per lattice site ∆ from gravity is 1137 Hz. Thelattice and Raman beams are all derived from a 1064 nm laser. Right: In situ atomiccloud width as a function of detuning of the two Raman beams. The 2D lattices wereset at 10ER = 20.3 kHz and VK = 2.3 kHz. The solid line is a fit to a lorentziancurve.
4.6.2 In Situ Measurements of Raman-Assisted Tunneling
We performed in situ measurements just as in the amplitude modulation tunneling
experiment, but now with Raman beams to induce Raman-assisted tunneling. We
acquired data by taking absorption images along the direction perpendiclar to the
2D lattice. First we scanned the detuning between the two Raman beams and looked
for expansion of the atom cloud. The two-dimensional lattices were both 10ER =
20.3 kHz with intensities in the Raman beams corresponding to VK = 2.3 kHz and
held for 400 ms. This is shown in the right of Fig. 4-10. A lorentzian fit gives
the resonance to be at 1214 ± 6 Hz and the width to be 54.6 ± 15.9 Hz where the
uncertainties are again 95% confidence levels, equivalent to two standard deviations.
Recall gravity should give a resonance at 1137 Hz. The discrepancy can be attributed
to slight misalignment of the laser beams and could be larger than in the amplitude
modulation case because we have one additional lattice beam, as well as two Raman
beams.
Once the resonance was found, we fixed the detuning of the Raman beams at
1214 Hz and looked at the atomic cloud width for different modulation times at
92
0 200 400 600 800 10004
6
8
10
12
14
Modulation Time (ms)
In−
Tra
p W
idth
(µm
)
0 0.4 0.8 1.2 1.6 2 2.4 2.8 3.2 3.6−1
0
1
2
3
4
5
6
7
VK/∆
In−
Tra
p E
xpan
sion
Rat
e (µ
m/s
)
Figure 4-11: Raman-assisted tunneling leads to linear expansion in time. The atomiccloud expansion as a function of time is plotted on the left for different Ramanintensities. On the right, the slope of the lines on the left plot are shown as a functionof Raman strength, and is fit to the square of J1(VK/∆). The Raman detuning wasfixed to the resonance value of 1214 Hz.
different Raman intensities. From our theoretical discussions, we should expect to
see a saturation in the tunneling as the Raman depth increases and even turn over
so that beyond a certain point, stronger drives lead to less tunneling per time. The
data are plotted in Fig. 4-11. We see a linear expansion in the width as a function
of time for all Raman intensities. We would expect the slope of this expansion to be
proportional to the tunneling amplitude. On the right of the figure I plot the slopes
as a function of Raman lattice strength. The line is a fit to the square of the Bessel
function, J1(VK/∆). Recall that for amplitude modulation, we saw an expansion of
the atomic cloud going as the square of time, whereas now the expansion goes linearly
with time.
We further explored this Bessel function behavior by looking at the atomic cloud
width as a function of Raman strength for fixed modulation time to observe the
high Raman beam intensity regime. This is shown in Fig. 4-12, where we plot the
obtained data, along with a fit to the square of J1(VK/∆) for data up to VK/∆ = 3.2
which is the value we had in Fig. 4-11. The fit agrees well with data up to the value
VK/∆ = 3.2, but does not fit the data well for higher Raman strengths. This could be
due to inhomogeneities in the tilt ∆ which gives rise to different resonance conditions
93
0 1 2 3 4 5 6 75
6
7
8
9
10
VK/∆
In−
Tra
p W
idth
(µm
)
Data
|J1(V
K/∆)|2
Figure 4-12: Raman-assisted tunneling for high Raman beam intensities. The Ramandetuning was fixed to the resonance value of 1214 Hz, and the Raman laser strengthwas varied. The tunneling restored by the running wave Raman beams qualitativelybehaves as a Bessel function as a function of the Raman strength. The line is a fit tothe square of J1(VK/∆) up to VK/∆ = 3.2 where the data in Fig. 4-11.
at different parts of the atom cloud or high intensity effects not captured in our model.
Further studies are needed to understand the discrepancy at high Raman intensities.
Also, the expansion behavior as the square of the Bessel function is not quite well
understood, although similar behavior has been observed by the Arimondo group
using phase modulation [92]. They attribute this behavior partly to interaction effects,
where by varying the atom number they can change the expansion behavior. Although
we also varied our atom number by more than a factor of 5, we were unable to observe
any difference in the expansion dynamics.
We also examined the cloud width for a given Raman strength and duration as
a function of lattice depth in the direction of the tilt. This is shown in Fig. 4-13.
The Raman detuning was fixed on resonance at 1214 Hz, the Raman strengths were
such that VK = 2.6 kHz and modulated for 400 ms. The curve is a fit of the analytic
94
5 6 7 8 9 10 11 12 136
8
10
12
14
16
Lattice Depth (ER)
In−
Tra
p W
idth
(µm
)
Figure 4-13: Raman-assisted tunneling as a function of the lattice depth. The curveis proportional to the analytic expression for the bare tunneling [29]. Contrast thiswith what we saw in the top right of Fig.4-8, where the width was proportional tothe square of the tunneling.
95
expression for tunneling up to a proportionality constant, and is of the form [29]
σ(VL) = CV3/4L e−2
√NL . (4.80)
This curve agrees well with the data. Compare this with the situation we saw with
amplitude modulation, where the width was increasing as the square of the bare
tunneling. Further studies will be needed to understand the cause of these different
expansion behaviors.
By scanning the detuning even further from the resonance at ∆, we have been
able to observe higher order tunneling processes. In particular we observed tunneling
at ∆/2 and possibly even lower fractions of ∆ in the left of Fig. 4-14. This data was
taken with the lattice in the tilt direction at 7ER, the perpendicular lattice at 10ER,
and the Raman intensity at 1.1 kHz for 400 ms. The ∆/2 resonance is clearly visible,
but the lower frequency resonances are not as sharp, possibly due to the fact that the
resonances are overlapping each other. A fit to three gaussians tells us that the left
most peak is centered at 348± 37 Hz with a width of 62± 41 Hz, the ∆/2 resonance
is centered at 560± 5 Hz with a width of 29± 5 Hz, and the ∆ resonance is centered
at 1122 ± 18 Hz with a width of 91 ± 24 Hz. The uncertainties are 95% confidence
levels, equivalent to two standard deviations.
We also observed tunneling at a Raman detuning of 2∆, which corresponds to
a two-photon Raman-assisted tunneling to next-nearest-neighbors. This is shown in
the right of Fig. 4-14, where the Raman beams were applied for 2 s but otherwise the
same experimental parameters as above. A fit to two gaussians tells us that the ∆
resonance is centered at 1137±31 Hz with a width of 107±31 Hz and the 2∆ resonance
is centered at 2262± 46 Hz with a width of 171± 87 Hz. Such higher order tunneling
processes, especially the ability to engineer a system where the next-nearest-neighbor
tunneing dominates the nearest-neighbor tunneling may be particularly attractive to
study systems with long range interactions.
96
0 200 400 600 800 1000 1200 1400 16005
6
7
8
9
10
Raman Detuning (Hz)
In−
Tra
p W
idth
(µm
)
600 1000 1400 1800 2200 26004
6
8
10
12
14
16
18
20
Raman Detuning (Hz)
In−
Tra
p W
idth
(µm
)
Figure 4-14: Higher-order and Next-nearest-neighbor tunneling with Raman beams.The left figure shows resonances at Raman detunings corresponding to ∆/2, whichare 4-photon processes to nearest neighbor sites, and even lower detunings. Theright figure shows a resonance at 2∆ which corresponds to 2-photon tunneling tonext-nearest-neighbor sites.
4.6.3 Time-of-Flight Measurements of Raman-Assisted Tun-
neling
Our successful observation of in situ Raman-assisted tunneling and the confirma-
tion of its Bessel function-like properties gives a strong indication that we have been
able to realize the Harper Hamiltonian. However, in order to confirm the creation
of the ground state of the Harper Hamiltonian, we need to be able to observe the
restored superfluid peaks whose spatial pattern should give an indication of the un-
derlying band structure which have been modified from the simple square lattice by
the position-dependent phase imprinted on the ground state wavefunction [82].
Unfortunately, we have so far been unable to observe the restoration of superfluid
peaks and the consequent novel band structure. Empirically, it appears that the cur-
vature of our laser beams may be preventing us from making the crucial observation.
Immediate future efforts will focus on understanding how we can establish superfluid
peaks in time-of-flight imaging. This should give the smoking-gun signature of the
ground state of the Harper Hamiltonian.
97
4.7 Conclusions
The technique of Raman-assisted tunneling allows one to study the Harper Hamil-
tonian with arbitrary flux. In addition, by including a third optical lattice, we can
significantly increase the interaction energy between the atoms such that we are in
the highly non-trivial regime of strongly correlated matter with artificial magnetic
fields, which can give rise to the celebrated Laughlin states of fractional quantum
Hall state fame [68]. Another direction is to add more spin states so that one can
study spin Hall effects which is a part of the burgeoning field of spintronics. Thus
there is significant promise in terms of the types of physics one can explore if one is
able to successfully create the Harper Hamiltonian and its ground state.
We have strong indications that the Harper Hamiltonian has been created with
Raman-assisted tunneling, as indicated by our in situ data which shows that our
tunneling behaves as a Bessel function. Thus we can say that we have confirmed the
absolute value of the Raman-assisted tunneling in the Harper Hamiltonian. However,
so far we have been unable to prove that we have created the ground state of this
Hamiltonian in time-of-flight images. There are indications that curvature of our laser
beams prevents us from achieving the ground state. We are able to restore coherence
in amplitude modulation, and the in situ dynamics of the atomic cloud is different
from that with Raman-assisted tunneling. Thus understanding the in situ dynamics
in further detail may give us a better understanding of the requirements to obtain
the ground state of the Harper Hamiltonian. Once this hurdle is over come and we
are able to observe the ground state of the Harper Hamiltonian, a whole new field of
exploring topological phases of matter with ultracold atoms in optical lattices will be
realized.
98
Chapter 5
Future Prospects
I have described in this thesis two different types of work. The Bragg scattering
work was geared toward the development of new measurement techniques to probe
new states of matter. On the other hand, the work on realizing the Harper Hamil-
tonian by employing Raman-assisted tunneling was geared toward directly creating
a new state of matter. Progress in the field of ultracold atoms will require both
development of probing techniques and creation of new matter, with both providing
inspiration to the other.
5.1 Prospects for Spin Hamiltonians
The Bragg scattering work was particularly geared towards its eventual use as a
probe of spin ordering, such as the antiferromagnetic state. The Heisenberg Hamil-
tonian which gives rise to magnetic states have an energy scale of J2/U , which is
on the order of 100 pK [15]. This is a very low energy scale, which at present only
adiabatic cooling techniques have any hope of directly reaching such low entropy
states [116, 76].
Spin gradient demagnetization cooling in its original form may very well take the
field down to the superexchange energy scale to realize spin Hamiltonians, but people,
including our group, have been thinking of other ways to access spin physics. One
notable success achieved by the group of Markus Greiner has been to use occupation
99
number of atoms on lattice sites as pseudo-spins to realize antiferromagnetic order-
ing [93]. Another approach, developed by our group, has been to use spin-dependent
lattices. In particular, one can create a low entropy effectively single component Mott
insulator where the spin-dependent lattice creates a 3D lattice for two spin species
interweaved between each other, and then adiabatically merge the two lattices. Be-
cause the initial ‘dual Mott’ system can have very low entropy, as long as the merging
is adiabatic, one should be able to achieve very low entropy spin-mixed states in a
3D lattice, which should be able to exhibit magnetic phases of matter as dictated by
the Heisenberg Hamiltonian.
Thus the future of spin Hamiltonians in ultracold atomic systems will most likely
involve some combination of brute force cooling techniques to reach picoKelvin tem-
peratures and ingeneous schemes which allow the manifestation of spin physics at
higher energy scales.
5.2 Prospects for Topological Phases of Matter
Topological phases of matter are usually related to magnetic fields. In neutral
cold atom systems, various techniques have already allowed the creation of effec-
tive magnetic fields. These include mechanical rotation of bulk quantum degenerate
gases [74, 1] and Raman processes in bulk quantum degenerate gases [72, 110, 17].
These techniques have the limitation that high effective magnetic fields, or more
specifically high effective magnetic flux per relevant spatial scale, are not easily re-
alizable. Lattice systems offer a promising route to realizing such effectively high
fluxes.
In the last few years, the group of Klaus Sengstock has realized the creation and
control of artificial gauge fields in a lattice by lattice modulation [101], the group of
Ian Spielman has achieved similar results by using Raman transitions and radiofre-
quency magnetic fields [60], and the group of Immanuel Bloch has realized staggered
fluxes by using a superlattice and Raman-assisted tunneling. The first two techniques
are only able to create position-independent phases, and the last is only able to cre-
100
ate staggered fluxes. It seems possible to extend all of these techniques to achieve
position-dependent, homogeneous fluxes, but the schemes are somewhat involved,
requiring more laser beams to address more transitions. If we are able to prove ex-
perimentally that we can create the ground state of the Harper Hamiltonian which
has non-trivial topological properties, it will be a much simpler way to realize the
same physics that the other groups are aiming for with a more complicated setup.
Once the Harper Hailtonian is realized, one can add interactions to the system
by using, for example, a third optical lattice, which should allow the study of frac-
tional quantum Hall states [94, 48] which manifest interesting topological properties
of matter.
Thus the general trend in the quest to realize artificial gauge fields and topological
phases of matter has been to add more and more laser beams, but a technique such
as ours which requires a relatively simple set up could hold the key to major progress
in the field.
5.3 Prospects for Quantum Simulation
The prospects for the program of quantum simulation seem bright [18]. It seems
that every few months a new type of system is realized in the laboratory, ranging
from graphene-like honeycomb lattice structures exhibiting Dirac points [104], ana-
logues of superconducting weak links with Bose-Einstein condensates in novel trap
geometries [117], and ‘Higgs’ type excitations in the superfluid to Mott insulator tran-
sition [31]. As can be seen in these examples, results from the field of ultracold atoms
is becoming strongly connected to concepts and ideas from condensed matter physics,
and even high energy physics.
Furthermore, different groups are also pushing the boundaries of ‘traditional’
ultracold atoms. By traditional, I mean alkali-metal atoms which have one va-
lence electron. Various groups have successfully created quantum degenerate gases
of two-valence-electron atoms such as calcium [66] strontium [98, 25], and ytter-
bium [102]. Even more exotic atoms have been cooled to quantum degeneracy, such
101
as chromium [43], erbium [4] and dysprosium [73], which all have large magnetic
dipole moments and therefore can give rise to anisotropic dipolar interactions. All
these different atomic systems will be able to explore a parameter space that alkali-
metal atoms have not yet been able to access thus far.
Numerous developments on all fronts of quantum simulation will certainly lead
to many new insights and discoveries of nature, most likely with implications be-
yond the field of ultracold atomic physics and further enriching and illuminating our
understanding of condensed matter systems and beyond.
102
Appendix A
Optical Lattice Calibration Using
Kapitza-Dirac Scattering
In this appendix, I describe the method we use to calibrate our optical lattice,
namely Kapitza-Dirac scattering [46].
The basic procedure is to pulse the optical lattice beams on for a very short time.
The short pulse then transfers momentum to the atoms, and then in a time-of-flight
absorption picture observe the spatial distribution of atoms. In particular the atoms
are scattered into discrete momentum states with probability given by how intense
the pulse was. Therefore by knowing how much optical power was in the pulses, and
measuring the population distribution of atoms in the different momentum states,
one can determine the resulting lattice depth created by the pulse.
The electric field E produced by two counter-propagating plane waves is given by
E(z, t) = E0f(t) sin(kz − ωt)e+ E0f(t) sin(kz + ωt)e (A.1)
= 2E0f(t) sin(kz) cos(ωt)e, (A.2)
where E0 is the electric field amplitude, k is the wavevector, ω is the angular frequency,
e is the polarization vector, and f(t) is the temporal envelope function.
Such a field will produce an AC Stark shift, leading to a standing wave potential
103
U(z, t) given by
U(z, t) =~ω2
R
δf 2(t) sin2(kz), (A.3)
where ωR is the single-photon Rabi frequency, given by ωR = µE0/~, where µ =
〈e|er|g〉 · e is the electric dipole matrix element connecting the ground (|g〉) and
excited (|e〉) states of the atom, and δ is the detuning. We may write
U0 ≡~ω2
R
δ. (A.4)
If the optical lattice is turned on for a very short time (i.e. τ 1/ωr, where ωr is the
recoil energy), the initial atomic wavefunction |Ψ0〉 will evolve in a way given by
|Ψ〉 = exp
− i~
∫dt′U(z, t′)
|Ψ0〉 (A.5)
= e−i2δω2Rτe
i2δω2Rτ cos(2kz)|Ψ0〉, (A.6)
where τ =∫dt′f 2(t′) and the integral is over the interaction duration. This short
time condition is called the Raman-Nath regime. For KD, our pulse is a constant
square pulse, so we let f(t′) = 1. We then make use of a Bessel function identity
eiα cosβ =∞∑
n=−∞
inJn(α)einβ (A.7)
to get
|Ψ〉 = e−i2δω2Rτ
∞∑n=−∞
inJn
(ω2
Rτ
2δ
)ei2nkz|Ψ0〉. (A.8)
We may let |Ψ0〉 = |g〉|0〉, where the first state represents the internal state of the
atoms and the second state represents the external state of the atoms. Note that we
104
can write
ei2nkz =∑p′
|p′ + 2n~k〉〈p′|, (A.9)
which can be used to get
∞∑n=−∞
ei2nkz|0〉 =∞∑
n=−∞
|2n~k〉. (A.10)
This can be inserted into Eq. A.8 to get
|Ψ〉 = e−i2δω2Rτ
∞∑n=−∞
inJn
(ω2
Rτ
2δ
)|g〉|2n~k〉. (A.11)
It is now easy to see that the probability that a state with momentum 2N~k will be
populated is given by
PN = J2N(θ) , N = 0,±1,±2, ..., where (A.12)
θ =ω2
Rτ
2δ. (A.13)
Our KD fit program takes as input the intensity of the KD peaks, which is pro-
portional to the probability PN of atoms being in a particular momentum state, and
the photodiode voltage V which is monitoring the power given to a particular lattice
beam. Thus we may rewrite Eq. A.12 as
PN = J2N(αV ), (A.14)
where θ = αV . The KD program determines the proportionality constant α, where
[α] =rad/V. Then we can write
αV =τ
2× ω2
R
δ. (A.15)
If we assume f(t) = 1 over the pulse duration (i.e. a square pulse), τ simply becomes
105
the pulse duration and is typically 12.5 µs with a rise and decay time of 5 ns.
Now, note that we can rewrite Eq. A.4 to get
ω2R
δ=U0
~. (A.16)
From atomic physics we know that
U0 ∝ ω2R ∝ I ∝ P ∝ V, (A.17)
where I is the intensity of the laser beam, P is the power incident on the photodiode,
and V is the voltage induced in the photodiode. From this we can write
U0 = γV, (A.18)
where [γ] =J/V and is a proportionality constant. We can insert this into Eq. A.16
to get
ω2R
δ=γV
~. (A.19)
Now we can insert this into Eq. A.15 to get
αV =τ
2× γV
~⇒ γ =
2~ατ. (A.20)
Now, we want recoil/V, which we call β, for calibration purposes. We have [Er] =
J/recoil, where Er = ~ωr, so
β =2~ατ× 1
Er
=2~ατ× 1
~ωr
=2α
τωr
. (A.21)
We can write ωr in terms of the wavelength λ of the lattice beam to make it easier
for plug and chug. Recall ~ωr = ~2k2/2m and k = 2π/λ where m is the mass of the
atom, so ωr = ~(2π/λ)2/2m. We can rewrite Eq. A.21 as
106
β =4αm
~τ×(λ
2π
)2
. (A.22)
107
108
Appendix B
Code to Construct Wannier
Functions
This appendix provides a copy of a MATLAB® code to construct maximally
localized Wannier functions for any lattice spacing and lattice depth that the user
can specify. In the process the band structure can also be extracted. A sample output
of this code for lattice spaing 532 nm and lattice depth 3ER is provided in Fig. B-1.
% Determine the Wannier function
% Written by Hirokazu Miyake
% Date: 4/21/2013
%
% lambda = lattice laser wavelength
% l = number of Fourier components to take
% Er = recoil energy
% V = V0/Er lattice depth in units of recoil energy
% q = q/k = q*a/pi where a = lattice spacing
clear;
lambda = 1064E-9; %lattice laser wavelength
dxn = 0.01;
x = lambda/2*(-10:dxn:10); %real space coordinates
dx = dxn*lambda/2; %discretization of space
x1 = 0; %center position of Wannier function
V = 3; %lattice depth in units of the recoil energy
l = 5;
wavevec = 2*pi/lambda; %wavevector
109
dn = 500;
qend = 1;
dq = 2*qend/dn;
H = zeros(2*l+1);
q = -qend:dq:qend;
nq = length(q);
Eeig = zeros(1,dn+1);
%Veig is c_j’s vs q
Veig = zeros(2*l+1,dn+1);
%construct the Hamiltonian and find eigenvalues and eigenvectors
%s is a sum over q
for s = 1:dn;
for n = 1:(2*l + 1)
for m = 1:(2*l + 1)
if n == m
H(n,m) = (2*(-l-1+n) + q(s)).^2 + V/2;
elseif abs(n - m) == 1
H(n,m) = -V/4;
else
H(n,m) = 0;
end
end
end
[W,D] = eig(H);
Eeig(s) = D(1,1);
for t = 1:(2*l+1)
Veig(t,s) = W(t,1);
end
end
nx = length(x);
wan1 = zeros(1,nx); %initialize wannier function
u=zeros(nq,nx); %u is q vs x
phi=zeros(nq,nx);
%create u_q(x)
% loop over q
for qi = 1:(nq-1)
for xi = 1:nx
%sum over j’s from -L to L
for k = 1:(2*l + 1)
u(qi,xi) = Veig(k,qi)*exp(1i*2*wavevec*x(xi)*(-l-1+k))...
+ u(qi,xi);
110
end
phi(qi,xi) = exp(1i*q(qi)*wavevec*x(xi)).*u(qi,xi);
end
end
%create phi_q(x)
%note that the phase has to be this specific one to create
%the maximally localized Wannier function
for qi = 1:(nq-1)
phi(qi,:) = exp(-1i*angle(phi(qi,x==0)))*phi(qi,:);
end
%create Wannier function w(x-x0)
%loop over x
for xi = 1:nx
%sum over q
for qi = 1:(nq-1)
wan1(xi) = phi(qi,xi)*exp(-1i*q(qi)*wavevec*x1) + wan1(xi);
end
end
%normalize Wannier function
wannorm1 = dx*sum(abs(wan1).^2);
wan1 = wan1/sqrt(wannorm1);
%plot Wannier function
figure(1);
plot(x/(lambda/2),real(wan1),’-k’,’linewidth’,3);
title (’Wannier Function’);
xlabel(’x/a’);
ylabel(’w(x)’);
111
−10 −5 0 5 10−500
0
500
1000
1500
2000Wannier Function
x/a
w(x
)
Figure B-1: Wannier function for lattice spacing 532 nm and lattice depth 3ER. Thex-axis is in units of the lattice spacing.
112
Appendix C
Bragg Scattering as a Probe of
Atomic Wave Functions and
Quantum Phase Transitions in
Optical Lattices
This appendix contains a reprint of Ref. [77]: Hirokazu Miyake, Georgios A.
Siviloglou, Graciana Puentes, David E. Pritchard, Wolfgang Ketterle, and David M.
Weld, Bragg Scattering as a Probe of Atomic Wave Functions and Quantum Phase
Transitions in Optical Lattices, Phys. Rev. Lett. 107, 175302 (2011).
113
Bragg Scattering as a Probe of Atomic Wave Functions and Quantum PhaseTransitions in Optical Lattices
Hirokazu Miyake, Georgios A. Siviloglou, Graciana Puentes, David E. Pritchard, Wolfgang Ketterle, and David M. Weld*
MIT-Harvard Center for Ultracold Atoms, Research Laboratory of Electronics, and Department of Physics,Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA
(Received 26 August 2011; published 21 October 2011)
We have observed Bragg scattering of photons from quantum degenerate 87Rb atoms in a three-
dimensional optical lattice. Bragg scattered light directly probes the microscopic crystal structure and
atomic wave function whose position and momentum width is Heisenberg limited. The spatial coherence
of the wave function leads to revivals in the Bragg scattered light due to the atomic Talbot effect. The
decay of revivals across the superfluid to Mott insulator transition indicates the loss of superfluid
coherence.
DOI: 10.1103/PhysRevLett.107.175302 PACS numbers: 67.85.d, 05.30.Rt, 37.10.Jk, 61.05.a
Ultracold atomic gases are an ideal system in whichto study many-body phenomena because of the relativeease with which parameters in the model Hamiltonian canbe tuned across a wide range [1,2]. Such studies haveresulted in a better understanding of various phasetransitions such as the Berezinskii-Kosterlitz-Thoulesstransition in two-dimensional systems [3], the BEC-BCScrossover of interacting fermions [4], and the superfluidto Mott insulator transition in a three-dimensional lattice[5]. One major goal of this field is to realize spin phasessuch as antiferromagnetic states to explore quantummagnetism and its interplay with high-temperature super-conductivity [6].
Concurrently, there are numerous efforts to developtechniques to probe and understand the atomic ensembleonce a new type of ordering is achieved. One recentdevelopment is the realization of single-site resolution ofatoms in two-dimensional optical lattices [7,8]. An alter-native method to measure in situ spatial ordering is thetechnique of Bragg scattering, often used in a condensedmatter context to determine crystal structure [9]. In par-ticular, Bragg scattering with neutrons led to the discoveryof antiferromagnetism in the solid state [10] and withx rays led to the discovery of the double helix structureof DNA [11].
Bragg scattering relies on the interference of wavesscattered coherently from an ensemble of particles. Inparticular, when atoms are arranged in a periodic patternin three spatial dimensions, there are certain angles of theincoming and reflected beams where scattering is dra-matically enhanced compared to other angles. This hasallowed crystallographers to use x rays to determinethe properties of crystals such as lattice geometry at theangstrom scale. We have applied this technique to ultra-cold atoms in a three-dimensional optical lattice by scat-tering near-infrared light where the atoms are spacedalmost 104 times farther apart than those in typical con-densed matter samples.
Pioneering works on Bragg scattering from cold atomsin optical lattices were done by the Hansch and Phillipsgroups by using laser cooled atoms [12]. These latticeswere sparsely populated, and the atoms occupied severalbands. In this Letter, we have used Bragg scattering tostudy bosonic atoms cooled to quantum degeneracy andplaced in a three-dimensional cubic lattice where the atomsoccupy the lowest band. In particular, we have measureddirectly the Heisenberg-limited width of both the positionand momentum of the single ground state atomic wavefunction in an optical lattice. Furthermore, there is a revivalof Bragg scattered light some time after releasing theatoms from the optical lattice, analogous to the opticalTalbot effect. This signal gives a direct measure of thecoherence of the superfluid component in the lattice.The experimental apparatus has been described in detail
elsewhere [13]. Briefly, laser cooling and evaporative cool-ing are used to achieve quantum degeneracy of 87Rb atomsin the jF ¼ 2; mF ¼ 2i state which are loaded into acrossed dipole trap whose trap frequencies range between30 and 160 Hz. Once quantum degeneracy is achieved, theoptical lattices generated by a single 1064 nm fiber laserare adiabatically ramped on. The lattices are calibrated byapplying a 12:5 s pulse and measuring the Kapitza-Diracdiffraction of the atoms and comparing this to theory. Thesystem typically contains about 105 atoms, which leads toup to 5 atoms per lattice site. The Bragg reflected light isdetected on a CCD camera which images along a directionwhich satisfies the Bragg condition.The Bragg scattering condition for a three-dimensional
cubic lattice is given by 2d sin ¼ np, where d is the
spacing between lattice planes, is the angle between theincoming beam and the lattice planes, n is any positiveinteger, and p is the wavelength of the probe beam. For
our experiment p is 780 nm and d is 532 nm. With these
conditions, the only allowed angle is 47 where n is oneand corresponds to Bragg reflection off the [1 0 0] plane orany equivalent plane. A schematic of the probe beam with
PRL 107, 175302 (2011) P HY S I CA L R EV I EW LE T T E R Sweek ending
21 OCTOBER 2011
0031-9007=11=107(17)=175302(5) 175302-1 2011 American Physical Society
respect to the atoms is shown in the inset in Fig. 1. Sincethe full angular width of the Bragg scattered light is small(measured to be 4:1 0:4), a precise alignment of theincident beam had to be performed at the 1=min repetitionrate of the experiment. In contrast, in one and two dimen-sions, diffractive light scattering occurs at any angle ofincidence, as recently shown with atoms in a two-dimensional optical lattice [14].
The probe beam at the Bragg angle had a typical powerof 0.3 mWand beam diameter of 300 m, large enough toilluminate the entire atomic cloud, and was detuned fromthe 52S1=2, F ¼ 2 ! 52P3=2, F
0 ¼ 3 cycling transition of87Rb by a few natural linewidths, where the natural line-width is 6 MHz. The detuning needs to be sufficient sothat the light traverses the entire atom cloud. Although theabsolute signal varies with the detuning, the conclusionsreached in this Letter appear not to depend strongly on theamplitude or sign of the detuning. The probe beam wasapplied for a few microseconds, enough to obtain a signalbut short enough to avoid heating effects.
As the lattice is increased, the Bragg reflected signalincreases as expected from the crystal ordering and tighterlocalization of the atoms (Fig. 1). For these data, the latticein the z direction, which is also the direction of gravity, wasramped to 15ER, where ER ¼ h2=ð2m2
LÞ is the recoilenergy, h is the Planck constant, m is the mass of 87Rb,and L ¼ 1064 nm is the lattice laser wavelength.Simultaneously, the lattices in the horizontal directionswere ramped to various lattice depths. Note that for ourgeometry the Bragg reflected intensity is insensitive tothe lattice depth in the z direction, because Bragg scat-tering occurs only in the horizontal xy plane, as wediscuss later.
The scattering of light by a collection of atoms can bemodeled in the following way. In the limit of low probeintensity and low optical density of the cloud, the Bornapproximation can be used. Then the scattering crosssection d=d can be written as a product of one-dimensional Debye-Waller factors where d=d /Q
i exp½ðriÞ2K2i =2, where i ¼ x; y; z is the index for
the three dimensions, Ki is the magnitude of the ith com-ponent of K, where K ¼ kin kout is the difference be-tween the incoming probe beam and Bragg scattered wave
vectors, and ri ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi@=ðm!iÞ
pis the harmonic oscillator
width, where @ is the reduced Planck constant and!i is thetrap frequency in the ith direction, which depends on theoptical lattice depth. The atoms are approximated asGaussian wave functions, which is a good approximationfor sufficiently large lattice depths. However, we find that,even for low lattice depths where significant superfluidcomponents are expected, this approximation describesthe Bragg scattered signal well.For our experimental conditions, Kz ¼ 0 because Bragg
scattering occurs in the horizontal xy plane. The latticedepths are controlled in a way such that they are the samein both the x and y directions which leads to x rx ¼ry. This simplifies the scattering cross section to
d=d / exp½ðxÞ2K2=2. Thus Bragg scattering al-lows the study of the spatial extent of the atomic wavefunction.The data can be compared to a no-free-parameter theo-
retical line, assuming noninteracting atoms. The Braggscattered intensity BðNLÞ as a function of lattice depthNL is proportional to exp½2
LK2=ð82
ffiffiffiffiffiffiffiNL
p Þ. Both theharmonic oscillator width x, which depends on the massof the rubidium atom and trap frequency, and change inwave vectorK are known. This theoretical line and the dataare shown in Fig. 1, and we see good agreement. Thus theBragg scattered light as a function of lattice depth can bewell described by a model where the atoms are assumed tobe noninteracting Gaussian atomic wave functions whosespatial width is determined by the lattice depth.Bragg scattering can also be used to probe the momen-
tum width of the atomic wave functions. This is done bymeasuring Bragg reflection as a function of the expansiontime of the atoms between a rapid lattice turn-off andBragg probe. In particular, the two horizontal latticeswere turned off from 15ER to 0ER in less than 1 s, andthe lattice in the z direction was kept at 15ER. Turning offthe lattices allows the atomic wave packets in each indi-vidual lattice well to expand freely in those directions. Thedata in Fig. 2 show that the signal has decayed in 30 s.The time it takes to lose crystal structure is roughly thetime it takes for the atoms to move over half of a latticedistance.The Debye-Waller factor can be used to determine more
precisely how the Bragg reflection should behave as afunction of the expansion time with no free parameters,
FIG. 1 (color online). Bragg scattering as a probe of the spatialwave function width. Bragg scattered intensity vs lattice depth inunits of the recoil energy. The right axis gives the correspondingroot-mean-square width of the wave function squared. The solidline is a no-free-parameter curve given by the Debye-Wallerfactor. Error bars are statistical. The inset is a schematic of thesetup for Bragg scattering.
PRL 107, 175302 (2011) P HY S I CA L R EV I EW LE T T E R Sweek ending
21 OCTOBER 2011
175302-2
assuming Gaussian atomic wave packets. This makesuse of the well-known time-dependent behavior of aHeisenberg-limited Gaussian wave packet in free spacewhich can be written as ½xðtÞ2 ¼ ðxÞ2 þ ðpÞ2t2=m2,where x and p are the uncertainty of position andmomentum, respectively, at the initial time and t is theexpansion time [15]. In a previous paper where the atomswere not quantum degenerate, the momentum distributionwas assumed to be determined by temperature [12]. Theresults in this Letter show that the momentum uncertaintyis Heisenberg limited, meaning ðxÞ2ðpÞ2 ¼ @
2=4, so theBragg scattered light BðtÞ as a function of expansion timedecays as exp½ðpÞ2K2t2=ð2m2Þ. This curve is alsoshown in Fig. 2. The curve is broadened by taking intoaccount the probe beam duration, which was 5 s andshows good agreement with the data. This analysis showsthat, by releasing the atoms from a lattice, we can directlyprobe the in situ time evolution of the ground state atomicwave functions with a spatial extent of tens of nanometers.Furthermore, the atomic spatial and momentum widths areseen to be Heisenberg limited.
Coherent many-body nonequilibrium dynamics can alsobe studied with Bragg scattering after a sudden release ofthe atoms from the optical lattice in three dimensions. Atlow lattice depths, the Bragg scattered signal shows reviv-als as a function of expansion time because of the rephas-ing of the superfluid atomic wave functions that wereoriginally confined in the lattices. Therefore, these revivalsignals should give a measure of the superfluid orderparameter. Furthermore, the revivals are analogous to theoptical Talbot effect, whose atomic version was observedpreviously for a thermal beam [16] and for a quantum
degenerate gas in the temporal domain [17]. Here weobserve the atomic Talbot effect for an interacting quantumdegenerate gas in a three-dimensional optical lattice.Revivals can be seen in Fig. 3 and become less pronouncedas the lattice depth increases.In particular, the characteristic revival time is deter-
mined by ¼ h=½@2ð2kÞ2=ð2mÞ and is 123 s for ourparameters, where 2k corresponds to the wave vector ofthe matter wave. However, with Bragg scattering we ex-pect the first revival at half that time of 61 s, which iswhat we observe in Fig. 3. This can be understood byrealizing that the atomic wave packets need only travelhalf the lattice spacing to constructively interfere with thewave packets traveling in the opposite directions from thenearest neighbor sites.The superfluid coherence as a function of lattice depth
can be studied by comparing the Bragg reflected signalwhen the atoms are in the optical lattice to the signal at thefirst revival. The revival signal as a function of lattice depthis shown in Fig. 4, where the superfluid to Mott insulatortransition is expected to occur around 13ER [5]. In a non-interacting system without any dissipation, one would ex-pect complete revivals. The decrease in revival fraction asthe lattice depth increases is consistent with the picturethat, as interactions increase, phase fluctuations amongneighboring lattice sites increase and, consequently, thesuperfluid fraction decreases. In the Mott insulating phase,revivals should completely disappear, except for a very
FIG. 2. Bragg scattering as a probe of the momentum wavefunction width. Bragg scattered intensity vs the free-expansiontime of the atoms after rapidly turning off the lattices from 15ER.The decay in signal indicates a melting of the crystal structure or,in other words, spreading of the atomic wave packet with a givenmomentum uncertainty. The gray area is a no-free-parametercurve using the Debye-Waller factor taking into account theprobe pulse duration of 5 s. Error bars are statistical.
FIG. 3 (color online). Revivals of Bragg scattered light. Braggscattered intensity vs the expansion time in three dimensions forthree different initial lattice depths: 5ER (circles), 8ER (tri-angles), and 15ER (squares). Each data set is normalized to theBragg intensity at the initial time. Revivals at lower latticedepths indicate coherence of the atoms across multiple latticesites. The lines are phenomenological fits to exponentiallydecaying sine waves. Different lattice depth data are offset forclarity, and the error bar is representative. The inset is a solutionto the one-dimensional Gross-Pitaevskii equation with experi-mental parameters and an initial state of a chain of Gaussianwave functions, showing the revivals of the density distributionas a function of time. The white dotted line represents the revivaltime.
PRL 107, 175302 (2011) P HY S I CA L R EV I EW LE T T E R Sweek ending
21 OCTOBER 2011
175302-3
weak signal at intermediate lattice depths due to particle-hole correlations [18].
To understand the measured revival decay, we havenumerically studied the one-dimensional Gross-Pitaevskiiequation that assumes interacting matter waves at zerotemperature. The simulations show that interactions andfinite size effects have a negligible effect on the decay ofrevivals. Empirically, randomizing the phase betweenneighboring sites reduces the revival fraction. The loss ofphase coherence across lattice sites could be due to factorssuch as temperature, beyond Gross-Pitaevskii equationeffects, or technical factors. We have considered quantumdepletion [19], but the calculated depletion fraction at 5ER
is 2%, too low to account for the observed decay. Futurework should provide a more complete picture of the decayof revivals.
Note that Bragg scattered revivals are complementary tothe observation of diffraction peaks in time-of-flight ab-sorption images [5]. Both are based on matter-wave inter-ference due to superfluid coherence: revivals at shortexpansion times and diffraction at long expansion times.Diffraction is a far-field effect possible only for finite sizesamples, whereas revivals are a near-field bulk effect. Theorder of the revival or the angular resolution of the diffrac-tion peaks determines whether these techniques probeshort-range or long-range spatial coherence. Further stud-ies are needed to determine which technique is moresensitive to certain aspects of the superfluid to Mott insu-lator transition.
In this Letter, we have not focused on the effects of theoccupation number on Bragg scattering. In principle, theatomic wave functions are more extended for higher occu-pation numbers, but this effect is small for our parameters.However, light scattering at higher occupation numberswill have an inelastic component due to colliding atomsor photoassociation [20]. The dependence of Bragg scat-tering on lattice depth suggests that these effects are notdominant.After Bragg scattering has been established as a probe
for Mott insulators, it can now be applied to study othertypes of quantum phases such as antiferromagnetic order-ing in both the occupation number and spin sectors [21,22].Although temperatures required to realize spin ordering areon the order of tens to hundreds of picokelvins, recentexperimental results suggest a way forward to lower thetemperature of a two-component Mott insulator [23].In conclusion, we have observed Bragg scattering of
near-resonant photons from a quantum degenerate Bosegas in a three-dimensional optical lattice. We have shownthat this technique probes not only the periodic structureof the atoms but also the spatial and momentum width ofthe localized atomic wave functions and the superfluidcoherence.We acknowledge Ian Spielman, Eugene Demler, Igor
Mekhov, Jesse Amato-Grill, Niklas Jepsen, and IvanaDimitrova for fruitful discussions and technical support.We acknowledge Yichao Yu for experimental assistance.We thank Jonathon Gillen for a critical reading of themanuscript. H.M. acknowledges support from theNDSEG program. This work was supported by the NSFthrough the Center of Ultracold Atoms, by NSF GrantNo. PHY-0969731, through an AFOSR MURI program,and under ARO Grant No. W911NF-07-1-0493 with fundsfrom the DARPA OLE program.
*Present address: Department of Physics, University of
California, Santa Barbara, CA 93106, USA.[1] M. Lewenstein et al., Adv. Phys. 56, 243 (2007).[2] I. Bloch, J. Dalibard, and W. Zwerger, Rev. Mod. Phys. 80,
885 (2008).[3] Z. Hadzibabic et al., Nature (London) 441, 1118 (2006).[4] M.W. Zwierlein et al., Nature (London) 435, 1047 (2005).[5] M. Greiner et al., Nature (London) 415, 39 (2002).[6] P. A. Lee, N. Nagaosa, and X.-G. Wen, Rev. Mod. Phys.
78, 17 (2006).[7] W. S. Bakr et al., Nature (London) 462, 74 (2009).[8] J. F. Sherson et al., Nature (London) 467, 68 (2010).[9] N.W. Ashcroft and N.D. Mermin, Solid State Physics
(Brooks Cole, Belmont, MA, 1976).[10] C. G. Shull and J. Samuel Smart, Phys. Rev. 76, 1256
(1949).[11] J. D. Watson and F. H. C. Crick, Nature (London) 171, 737
(1953).
4 6 8 10 12 14 16
0
50
100
150
200
250
Rev
ival
Inte
nsity
(ar
b.)
4 6 8 10 12 14 16
0
0.2
0.4
0.6
0.8
1
Lattice Depth (ER
)
Nor
m. R
eviv
al In
tens
ity (
arb.
)
FIG. 4. Bragg scattering revival as a probe of superfluid co-herence. Top: Bragg scattered revival signal at 60 s vs latticedepth. Bottom: The top plot normalized to the Bragg scatteredsignal without expansion. We see a decrease in revival, andconsequently superfluid coherence, as the lattice is increased.Error bars are statistical.
PRL 107, 175302 (2011) P HY S I CA L R EV I EW LE T T E R Sweek ending
21 OCTOBER 2011
175302-4
[12] M. Weidemuller et al., Phys. Rev. Lett. 75, 4583 (1995);G. Birkl et al., Phys. Rev. Lett. 75, 2823 (1995); G. Raithelet al., Phys. Rev. Lett. 78, 630 (1997); M. Weidemuller, A.Gorlitz, T.W. Hansch, and A. Hemmerich, Phys. Rev. A58, 4647 (1998).
[13] E.W. Streed et al., Rev. Sci. Instrum. 77, 023106 (2006).[14] C. Weitenberg et al., Phys. Rev. Lett. 106, 215301 (2011).[15] L. I. Schiff, Quantum Mechanics (McGraw-Hill, New
York, 1968).
[16] M. S. Chapman et al., Phys. Rev. A 51, R14 (1995).[17] L. Deng et al., Phys. Rev. Lett. 83, 5407 (1999).[18] F. Gerbier et al., Phys. Rev. Lett. 95, 050404 (2005).[19] K. Xu et al., Phys. Rev. Lett. 96, 180405 (2006).[20] A. Gallagher and D. E. Pritchard, Phys. Rev. Lett. 63, 957
(1989).[21] T.A. Corcovilos et al., Phys. Rev. A 81, 013415 (2010).[22] J. Simon et al., Nature (London) 472, 307 (2011).[23] P. Medley et al., Phys. Rev. Lett. 106, 195301 (2011).
PRL 107, 175302 (2011) P HY S I CA L R EV I EW LE T T E R Sweek ending
21 OCTOBER 2011
175302-5
Appendix D
Spin Gradient Demagnetization
Cooling of Ultracold Atoms
This appendix contains a reprint of Ref. [76]: Patrick Medley, David M. Weld,
Hirokazu Miyake, David E. Pritchard, and Wolfgang Ketterle, Spin Gradient Demag-
netization Cooling of Ultracold Atoms, Phys. Rev. Lett. 106, 195301 (2011).
119
Spin Gradient Demagnetization Cooling of Ultracold Atoms
Patrick Medley,* David M. Weld,† Hirokazu Miyake, David E. Pritchard, and Wolfgang Ketterle
MIT-Harvard Center for Ultracold Atoms, Research Laboratory of Electronics, and Department of Physics,Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA
(Received 12 January 2011; revised manuscript received 16 March 2011; published 12 May 2011)
We demonstrate a new cooling method in which a time-varying magnetic field gradient is applied to an
ultracold spin mixture. This enables preparation of isolated spin distributions at positive and negative
effective spin temperatures of50 pK. The spin system can also be used to cool other degrees of freedom,
and we have used this coupling to cool an apparently equilibrated Mott insulator of rubidium atoms to
350 pK. These are the lowest temperatures ever measured in any system. The entropy of the spin mixture
is in the regime where magnetic ordering is expected.
DOI: 10.1103/PhysRevLett.106.195301 PACS numbers: 67.85.d, 03.75.Mn, 05.30.Jp, 75.30.Sg
Attainment of lower temperatures has often enableddiscovery of new phenomena, from superconductivity toBose-Einstein condensation. Currently, there is much in-terest in the possibility of observing correlated magneticquantum phases in lattice-trapped ultracold atoms [1–3].The relevant critical temperatures are on the order of200 pK, lower than any previously achieved temperature[4,5]. Realization of this temperature scale requires thedevelopment of new methods of refrigeration which canbe applied to ultracold atoms. Many such techniques havebeen proposed [6–13] but await experimental realization.The cooling method we demonstrate here opens up apreviously inaccessible temperature regime and providesa realistic path to the observation of magnetic quantumphase transitions in optical lattices.
The new cooling method is applied to an opticallytrapped cloud of cold atoms in a mixture of two internalstates with different magnetic moments [14]. Applicationof a strong magnetic field gradient results in almost com-plete spatial segregation of the two spin components. The‘‘mixed region’’ of spins between the pure-spin domainshas a width which is proportional to the temperature T andinversely proportional to the applied gradient. Reducingthe gradient mixes the two components, and due to themixing entropy the temperature is dramatically reduced.Since Mott insulators can be prepared with an entropy perparticle much lower than kB ln2, our scheme is a practicalway of creating a low entropy mixture in the regime wheremagnetic ordering is expected. This cooling scheme intro-duces several new concepts. It implements demagnetiza-tion cooling with spin transport instead of spin flips, allowsdecoupling of the spin and kinetic temperatures, and en-ables the realization of negative spin temperatures. Longtunneling times in the lattice allow two different imple-mentations of our cooling scheme.
When the gradient is changed faster than the spin re-laxation rate, the spin system is effectively isolated from allother degrees of freedom, and very low spin temperaturescan be achieved. Contrastingly, when the gradient is
changed slowly enough, the spin system is fully equili-brated and can absorb entropy from other degrees of free-dom, cooling the whole sample. As shown in Fig. 1,reduction of the gradient after (before) the optical latticehas been raised realizes the regime of isolated (equili-brated) spins.First, we discuss isolated spins, of which atoms in a Mott
insulating state are an almost ideal realization. Spin dis-tributions relax by two atoms exchanging sites through asecond-order tunneling process. The time scale of thisrelaxation is typically 1 s, and the gradient can easily bevaried much faster. The equilibrium spin distribution de-pends only on the ratio of the applied gradient rjBj andtemperature T. When rjBj is changed, the effective tem-perature of the decoupled spin degrees of freedom (spintemperature) is rescaled proportionally. This enables therealization of spin distributions with a very low positive(or, if the sign of the gradient is changed, negative) effec-tive temperature. Negative temperatures can occur only forsystems with an upper bound on the energy and correspond
FIG. 1. Two different cooling protocols realizing the cases ofisolated and equilibrated spin systems. (a) Experimental ‘‘phasediagram’’ of lattice depth vs applied gradient. Dashed lines showtwo different paths which connect the high-gradient superfluidstate and the low-gradient Mott insulating state. (b) and (c) showthe lattice depth (solid line) and gradient strength (dashed line)vs time for the two cases (equilibrated spins and isolated spins,respectively) in (a). The shape of the lattice ramp-up is designedto ensure maximum equilibration.
PRL 106, 195301 (2011) P HY S I CA L R EV I EW LE T T E R Sweek ending13 MAY 2011
0031-9007=11=106(19)=195301(4) 195301-1 2011 American Physical Society
to an inverted Boltzmann distribution with the largestpopulation in the highest-energy state [15].
Figure 2 shows the results of experiments on isolatedspins. Fits to data on equilibrated spins indicate an initialtemperature of 6.3 nK (see Figs. 3 and 4). While reductionof the gradient by a factor of 1000 would be expected toreduce the effective temperature of the isolated spins to6.3 pK, our finite optical resolution allows us only to assertan upper bound of 50 20 pK. We have held the spins forup to several seconds in the lattice. No heating is observedduring a 1-s hold, but after 3 s the 50 pK distribution isobserved to heat to about 70 pK. Similarly, at small nega-tive gradients, we observe a negative-temperature distribu-tion with a temperature closer to zero than 50 20 pK.Since the total energy is monotonic in 1=T, these areamong the most extreme thermodynamic states ever mea-sured in any system [16,17]. Very low condensate releaseenergies (not temperatures) have been observed previously[18]. Note that the spin temperatures we report are muchlower than those attainable by magnetic trapping or opti-cally pumping a system into a single spin state: Even for afractional population of 106 in other spin states causedby imperfect pumping or spin-flip collisions [19], the spintemperature in a bias field of 100 mG is 500 nK, assuminga magnetic moment of one Bohr magneton. In our experi-ment, the energy scale is set by the product of the Bohrmagneton with the magnetic field gradient and the latticespacing, and it is the relative ease of achieving very smallvalues of this parameter (corresponding to B times a fieldof less than 1 G) which allows us to reach such low spintemperatures.
The mixed-spin region comprises a spectrum of spinexcitations, the energy of which can be tuned by adjustingthe strength of the gradient [20]. A spin excitation consistsof a spin-up atom swapping places with a spin-down atomon the other side of the zero-temperature spin domainboundary. Reduction of the gradient reduces the energyof spin excitations. In the regime of equilibrated spins, thiscauses entropy to flow into the mixed-spin region fromother degrees of freedom. This lowers the temperature
of the whole system in a manner locally analogous tostandard single-shot adiabatic demagnetization refrigera-tion [21,22]. The kinetic excitations of a trapped Mottinsulator are particle-hole excitations [4,23]. On a micro-scopic level, particle-hole excitations can couple directlyto spin excitations if a particle and a hole on opposite sidesof the spin domain boundary annihilate. Energy may beabsorbed by quasiparticles in the first band or by the spinexcitations themselves [24]. In the superfluid phase andduring lattice ramp-up, the kinetic excitations are differentand entropy transfer is expected to occur faster and in morecomplex ways. Experimentally, it is easier to maintainadiabaticity if most of the path along which the gradientis changed is in a regime of fast relaxation times [e.g., thelower path in Fig. 1(a)]. Thermodynamically, this adiabaticcooling method is a redistribution of entropy from thekinetic degrees of freedom to the entropy which resultsfrom partially mixing the two initially segregated spindomains.It is easily possible for the mixed region to absorb nearly
all of the entropy of the system. In a one-componentharmonically trapped Mott insulator which is at a tempera-ture low enough for the particle-hole approximation tohold, the approximate total entropy is kB lnð2Þ times thevolume of the ‘‘shells’’ between the Mott plateaux [4]. Themaximum entropy of the mixed region is realized when, atlow gradient, it is broadened to a substantial fraction of thetotal size. In that situation, the entropy per site approacheskB lnðnþ 1Þ, where n is the local number of indistinguish-able bosons per site (in our samples, n varies across the trap
FIG. 2. Preparation of low positive and negative spin tempera-tures. Measured spin temperature vs final gradient, for the caseof isolated spins. Error bars are statistical.
FIG. 3. Entropy transfer from other degrees of freedom to thespin system. Circles represent measured width of the mixedregion vs final magnetic field gradient, for the ‘‘equilibratedspins’’ protocol (see Fig. 1). Error bars are statistical. The dashedline represents the expected behavior assuming no cooling. Thedash-dotted line shows the minimum resolvable width. The solidcurve is the theoretical prediction, assuming an initial tempera-ture of 6.3 nK and including the effects of optical resolution (seeRef. [25] for details). Insets show spin images at the indicatedpoints, the corresponding vertically integrated spin profiles(squares), and the fit to the expected form of a tanh functiontimes the overall density distribution (solid line; see Ref. [20] fordetails). Axis units and gray scales in the two insets are arbitrarybut identical.
PRL 106, 195301 (2011) P HY S I CA L R EV I EW LE T T E R Sweek ending13 MAY 2011
195301-2
between 1 and 3). Themaximum entropy of the spin systemis thus larger than the entropy of the kinetic degrees offreedom. Thus, substantial cooling of the system can beachieved with a single gradient demagnetization ramp.We have made a more quantitative analysis of spingradient demagnetization by calculating entropy-versus-temperature curves of our system in various field gradients[25]. The results confirm the qualitative argument aboveand show that spin gradient demagnetization cooling iscapable of cooling well below the expected Curietemperature.
Figures 3 and 4 show the results of spin gradient demag-netization cooling experiments. As the gradient is reduced,the width of the domain wall increases (see Fig. 3), in-dicating transfer of entropy from the kinetic degrees offreedom to the spins. The width increases much lesssteeply than would be expected for an isothermal sample,implying cooling. The observed domain wall width can beconverted to a temperature using spin gradient thermom-etry. The measured temperature falls rapidly as the gradientis lowered (see Fig. 4). The lowest measured temperature is350 50 pK, making this the coldest implementation ofthe atomic Mott insulator. An important goal for futureexperiments will be to develop an additional method ofthermometry to measure the temperature of the kineticdegrees of freedom at this very low temperature scale(for progress towards this goal, see Refs. [26–28]). Thistemperature is colder than the lowest temperature evermeasured in an equilibrated kinetic system [29], and it iswithin a factor of 2 of the expected magnetic orderingtemperature [5].
Theoretical curves in Figs. 3 and 4 (see also Ref. [25])show reasonable agreement with the data. These curves
were fitted to the measured temperatures by using only onefree parameter: the initial temperature at the maximumgradient. The initial temperature inferred from this fit is6.3 nK. In our earlier work on thermometry [20], where thelowest measured temperature was 1 nK, some adiabaticdemagnetization cooling may have occurred during thepreparation of the system. The flattening out observed inthe measured data at low gradients could be a signal that allaccessible entropy has been pumped into the spin system.There are both practical and theoretical limits on the
temperatures which can be attained with spin gradientdemagnetization cooling. In traditional magnetic refrigera-tion experiments, the minimum temperature is often set bythe minimum achievable magnetic field or the presence ofinternal fields in the refrigerant [30]. Analogues of boththese limits are relevant here. Practically, the ratio betweenthe highest and lowest attainable magnetic field gradientsis an upper bound on the ratio between the initial and finaltemperatures. In our experiment, the maximum value ofrjBij=rjBfj is about 1000 (limited mainly by the accu-
racy of determining the zero crossing of the gradient),which would give a minimum temperature below 10 pK.Another limit stems from the small difference between theinterspin interaction energy U"# and the mean of the intra-
spin interaction energies ðU"" þU##Þ=2. For 87Rb, this limit
is not expected to preclude cooling below the expectedmagnetic ordering temperature [25]. The fundamentallimit is set by the total entropy of the system. A singleMott insulator has an entropy per particle much less thankB ln2. At high gradient, the spin entropy is negligible.Lowering the gradient provides a controlled way of creat-ing a completely mixed two-component system at the sameentropy which is low enough for spin ordering to occur.This technique thus provides a specific and realisticmethod of realizing magnetic phase transitions in opticallattices.Our scheme differs from single-shot adiabatic demag-
netization refrigeration (including that demonstrated in agas of chromium atoms [31]) in that the magnetic field isreplaced by a magnetic field gradient and spin-flip colli-sions by spin transport. The chromium scheme cannot beapplied to alkali atoms due to the much slower spin-fliprates, and extending it from microkelvins to picokelvinswould require submicrogauss magnetic field control. Inprevious work [20], we suggested that adiabatic reductionof the gradient could be used for cooling, and some aspectsof this proposal have recently been theoretically addressedand verified [32].The concept behind spin gradient demagnetization cool-
ing is compelling; if adiabaticity can be maintained, thenstrong cooling in a lattice will occur. Our experimentalimplementation was designed to allow the system to equili-brate as much as possible at low lattice depths. We havetested reversibility by replacing the single gradient ramp bya sequence of ramp down, ramp up, ramp down. This led to
FIG. 4. Spin gradient demagnetization cooling. Circles repre-sent measured temperature vs final magnetic field gradient, forthe equilibrated spins protocol (see Fig. 1). Error bars arestatistical. These measurements are the same as those shownin Fig. 3. The dashed line follows the isothermal trajectory, andthe dash-dotted line shows the resolution limit. The solid line isthe theoretical prediction, assuming an initial temperature of6.3 nK and including the effects of optical resolution. The dottedline is the theoretical prediction without the effects of opticalresolution.
PRL 106, 195301 (2011) P HY S I CA L R EV I EW LE T T E R Sweek ending13 MAY 2011
195301-3
no detectable difference in the final measured temperature.This indicates that the gradient ramps are adiabatic.Equilibration in the Mott insulator is more difficult todemonstrate, although the previously demonstrated agree-ment between spin gradient thermometry and cloud sizethermometry at high temperatures [20] indicates that thekinetic and spin degrees of freedom are equilibrated in thatregime. The fact that the spin distribution fits well to theform expected of an equilibrated spin system is alsoevidence for equilibration, as is the one-parameter fit toour theoretical predictions (which assume adiabaticity).However, if the lattice is deepened, then lowered to zero,and then raised again, heating is generally observed.Thus, we cannot rule out the existence of long-lived meta-stable excitations in the Mott insulating state which do notcouple to the spin degrees of freedom and thus do notinfluence our temperature measurement. Other experi-ments have seen evidence of long equilibration times inthe Mott insulator [24,33]. For quicker equilibration, spingradient demagnetization cooling could be implementedwith lighter atomic species (e.g., 7Li or 4He) and/orshorter period optical lattices.
The cooling technique using isolated spins presentedhere has achieved spin temperatures and entropies wellbelow the critical values for magnetic ordering, and spingradient demagnetization cooling of equilibrated spins hascooled to a point within reach of the critical values. Thiswork thus opens a realistic path towards observation ofsuperexchange-driven phase transitions in optical latticesand extends the potential of ultracold atoms trapped inoptical lattices to be used as flexible quantum simulatorsof strongly interacting many-body systems.
We acknowledge discussions with Eugene Demler,Takuya Kitagawa, David Pekker, and Aditi Mitra. Wethank Aviv Keshet for a critical reading of the manuscript.H.M. acknowledges support from the NDSEG program.This work was supported by the NSF, through a MURIprogram, and under ARO Grant No. W911NF-07-1-0493with funds from the DARPA OLE program.
*Present address: Department of Physics, StanfordUniversity, Stanford, CA 94305, USA.†[email protected]
[1] M. Lewenstein et al., Adv. Phys. 56, 243 (2007).[2] I. Bloch, J. Dalibard, and W. Zwerger, Rev. Mod. Phys. 80,
885 (2008).
[3] L.-M. Duan, E. Demler, and M.D. Lukin, Phys. Rev. Lett.91, 090402 (2003).
[4] T.-L. Ho and Q. Zhou, Phys. Rev. Lett. 99, 120404(2007).
[5] B. Capogrosso-Sansone, S. G. Soyler, N.V. Prokof’ev, andB.V. Svistunov, Phys. Rev. A 81, 053622 (2010).
[6] F. Werner, O. Parcollet, A. Georges, and S. R. Hassan,Phys. Rev. Lett. 95, 056401 (2005).
[7] M. Popp, J.-J. Garcia-Ripoll, K.G. Vollbrecht, and J. I.Cirac, Phys. Rev. A 74, 013622 (2006).
[8] B. Capogrosso-Sansone, S. G. Soyler, N. Prokof’ev, andB. Svistunov, Phys. Rev. A 77, 015602 (2008).
[9] P. Rabl, A. J. Daley, P. O. Fedichev, J. I. Cirac, and P.Zoller, Phys. Rev. Lett. 91, 110403 (2003).
[10] J.-S. Bernier et al., Phys. Rev. A 79, 061601 (2009).[11] T.-L. Ho and Q. Zhou, Proc. Natl. Acad. Sci. U.S.A. 106,
6916 (2009).[12] T.-L. Ho and Q. Zhou, arXiv:0911.5506.[13] J. Catani et al., Phys. Rev. Lett. 103, 140401 (2009).[14] See supplemental material at http://link.aps.org/
supplemental/10.1103/PhysRevLett.106.195301 for de-tailed experimental methods.
[15] F. Reif, Fundamentals of Statistical and Thermal Physics(McGraw-Hill, New York, 1965).
[16] P. J. Hakonen, R. T. Vuorinen, and J. E. Martikainen, Phys.Rev. Lett. 70, 2818 (1993).
[17] J. T. Tuoriniemi et al., Phys. Rev. Lett. 84, 370 (2000).[18] T. Kraemer et al., Appl. Phys. B 79, 1013 (2004).[19] A part-per-million spin impurity level would, for example,
be present in a 100 Hz magnetic trap with a 100-s lifetimedue to dipolar relaxation.
[20] D.M. Weld et al., Phys. Rev. Lett. 103, 245301(2009).
[21] W. F. Giauque, J. Am. Chem. Soc. 49, 1864 (1927).[22] P. Debye, Ann. Phys. (Leipzig) 386, 1154 (1926).[23] F. Gerbier, Phys. Rev. Lett. 99, 120405 (2007).[24] R. Sensarma et al., Phys. Rev. B 82, 224302 (2010).[25] D.M. Weld, H. Miyake, P. Medley, D. E. Pritchard, and
W. Ketterle, Phys. Rev. A 82, 051603(R) (2010).[26] W. S. Bakr et al., Science 329, 547 (2010).[27] J. F. Sherson et al., Nature (London) 467, 68 (2010).[28] N. Gemelke, X. Zhang, C.-L. Hung, and C. Chin, Nature
(London) 460, 995 (2009).[29] A. E. Leanhardt et al., Science 301, 1513 (2003).[30] A. S. Oja and O.V. Lounasmaa, Rev. Mod. Phys. 69, 1
(1997).[31] M. Fattori et al., Nature Phys. 2, 765 (2006).[32] S. S. Natu and E. J. Mueller, Phys. Rev. A 82, 013612
(2010).[33] C.-L. Hung, X. Zhang, N. Gemelke, and C. Chin, Phys.
Rev. Lett. 104, 160403 (2010).
PRL 106, 195301 (2011) P HY S I CA L R EV I EW LE T T E R Sweek ending13 MAY 2011
195301-4
124
Appendix E
Thermometry and refrigeration in
a two-component Mott insulator of
ultracold atoms
This appendix contains a reprint of Ref. [116]: David M. Weld, Hirokazu Miyake,
Patrick Medley, David E. Pritchard, and Wolfgang Ketterle, Thermometry and re-
frigeration in a two-component Mott insulator of ultracold atoms, Phys. Rev. A 82,
051603(R) (2010).
125
RAPID COMMUNICATIONS
PHYSICAL REVIEW A 82, 051603(R) (2010)
Thermometry and refrigeration in a two-component Mott insulator of ultracold atoms
David M. Weld,* Hirokazu Miyake, Patrick Medley,† David E. Pritchard, and Wolfgang KetterleMIT-Harvard Center for Ultracold Atoms, Research Laboratory of Electronics, and Department of Physics,
Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA(Received 1 September 2010; published 18 November 2010)
Interesting spin Hamiltonians can be realized with ultracold atoms in a two-component Mott insulator (2CMI)[Adv. Phys. 56, 243 (2007); Rev. Mod. Phys. 80, 885 (2008)]. It was recently demonstrated that the applicationof a magnetic field gradient to the 2CMI enables new techniques of thermometry [Phys. Rev. Lett. 103, 245301(2009)] and adiabatic cooling [e-print arXiv:1006.4674]. Here we present a theoretical description which providesquantitative analysis of these two techniques. We show that adiabatic reduction of the field gradient is capable ofcooling below the Curie or Neel temperature of certain spin-ordered phases.
DOI: 10.1103/PhysRevA.82.051603 PACS number(s): 03.75.Mn, 37.10.Jk, 05.30.Jp, 37.10.De
The possibility of using ultracold lattice-trapped gases asgeneral simulators of strongly interacting many-body systemshas excited increasing interest in recent years [1–3]. SpinHamiltonians are a natural candidate for quantum simulation,especially given the relevance of doped antiferromagneticsystems to the important open problem of high-Tc supercon-ductivity [4]. The 2CMI is the starting point for the simulationof electronic spin systems in lattices [5–9]. Spin-exchange-stabilized magnetically ordered states are expected to exist inthe 2CMI [10], and observation of these states and transitionsbetween them would open up an exciting new field at theintersection of atomic and condensed matter physics. Themain obstacle which has so far prevented the observation ofspin-ordered states in the 2CMI is the very low temperaturescale required for spin ordering [11]. Quantum Monte Carlocalculations have predicted Curie and Neel temperatures on theorder of 200 pK for the ferromagnetic and antiferromagneticstates of 87Rb in a 532-nm lattice [12]. This is a lower temper-ature than has ever been measured in any system. Clearly, newmethods of thermometry and refrigeration are required.
The recently demonstrated technique of spin gradientthermometry [13] should allow measurement of temperaturesdown to the spin exchange scale in the 2CMI. The relatedmethod of spin gradient demagnetization cooling is capable ofcooling to the neighborhood of the critical temperature for spinordering [14]. Together, these new techniques open a realisticprospect of preparing spin-ordered phases in the 2CMI. Inorder to compare experimental results with theory, we havedeveloped a simple theoretical model of the 2CMI and used itto calculate the expected response of our system to spin gradi-ent thermometry and spin gradient demagnetization cooling.
Our treatment of the 2CMI is similar in approach to thestudies of cooling in the one-component Mott insulator pre-sented in Refs. [11] and [15] in that it is based on calculationsof entropy-versus-temperature curves for various values ofcontrol parameters. Our model neglects the effects of tunnelingand treats each lattice site separately, yet is capable of qualita-tively reproducing observed cooling curves using only one fit
*[email protected]†Present address: Department of Physics, Stanford University,
Stanford, CA 94305, USA.
parameter (the initial temperature) [14]. Our results thus com-plement, and are in qualitative agreement with, the classicalmean-field and Monte Carlo analysis of Natu and Mueller [16].
The inputs to the calculation are the measured trapfrequencies (ωx,ωy,ωz), the total atom number N , and theapplied magnetic field gradient ∇|B| (along with various fixedparameters like the scattering lengths and magnetic momentof the atoms and the lattice constant). This allows directcomparison with experiment. The particular trap frequenciesassumed here are 2π × (40,156,141) Hz. We assume anatom number of 17 000, leading to an occupation numberof 3 in the center of the cloud. These values were chosenbecause they are typical in our experiments. The scatteringlengths we assumed are a↑↑ = 100.4a0, a↓↓ = 98.98a0, anda↑↓ = 98.98a0, where a0 is the Bohr radius and states ↑ and ↓are the |F = 1,mF = −1〉 and |F = 2,mF = −2〉 hyperfinestates of 87Rb; these values represent the results of the mostrecent theoretical calculations available [17].
Detailed technical descriptions of spin gradient thermom-etry and spin gradient demagnetization cooling are presentedin Refs. [13] and [14], respectively. Both techniques are basedon the 2CMI in a magnetic field gradient. Since the twocomponents have different magnetic moments, the gradientpulls them toward opposite sides of the trap, creating two spindomains which remain in thermal contact. At zero temperature,there will be a zero-width boundary between the two domains,but at finite temperature a mixed region composed of spinexcitations will be present.
Since the total magnetization is always chosen to be zero,the average value of the magnetic field is canceled by aLagrange multiplier and can be subtracted from the real fieldB(x). This allows us to write the field as Beff = ∇|B| · xi ,where xi is the vector from the trap center to lattice site i
projected along the direction of the magnetic field gradient.Note that Beff = 0 at the trap center. If tunneling is neglected,then, at a magnetic field gradient ∇|B|, the energy of aconfiguration with n↑ up spins and n↓ down spins at latticesite i is
Ei(n↑,n↓,∇|B|) = p∇|B| · xi(n↑ − n↓)
+ 1
2
∑
σ
Uσσnσ (nσ − 1) + U↑↓n↑n↓
+Vi(n↑ + n↓) − µ↑n↑ − µ↓n↓, (1)
1050-2947/2010/82(5)/051603(4) 051603-1 ©2010 The American Physical Society
RAPID COMMUNICATIONS
WELD, MIYAKE, MEDLEY, PRITCHARD, AND KETTERLE PHYSICAL REVIEW A 82, 051603(R) (2010)
where p is the amplitude of the effective magnetic moment ofthe atoms, xi = |xi |, σ = ↑,↓, Uab is the interaction energybetween spin a and spin b, Vi = (m/2)(ω2
xx2i + ω2
yy2i + ω2
zz2i )
is the optical trapping potential at site i, yi and zi are thedistances of site i from the trap center in the two directionstransverse to the gradient, and µa is the chemical potential ofspin a. The chemical potential is set by the requirement thatthe number of atoms of each spin be equal to half the totalexperimentally measured number.
Equation (1) can be used in the grand canonical ensemble toinfer the thermal probability of different occupation numbersof the two spins. The partition function at lattice site i
is Zi(∇|B|) = ∑n↑,n↓ exp[−Ei(n↑,n↓,∇|B|)/kBT ], where
kB is Boltzmann’s constant, T is the temperature, and thesummation is over all possible combinations of n↑ and n↓ (eachcombination is counted only once, due to indistinguishabilityof the atoms). The probability of having n↑ up spins and n↓down spins at lattice site i is then
pi(n↑,n↓,∇|B|,T ) = exp[−Ei(n↑,n↓,∇|B|)/kBT ]
Zi
, (2)
and the resulting entropy at site i is
Si(∇|B|,T ) = −∑
n↑,n↓pi ln pi, (3)
where the summation is performed in the same way as for thepartition function. The only additional approximation neededis a truncation of the sums over spin configurations. For ourexperimental parameters, configurations corresponding to atotal atom number per site n greater than 4 can be neglected,and we have truncated the sums accordingly. This truncationis reminiscent of, but more general than, the particle-holeapproximation [11,15]. The site entropy Si of Eq. (3) issummed over all lattice sites to extract the total entropyas a function of temperature and field gradient. From thisoutput one can extract column-integrated images (Fig. 1),entropy-versus-temperature curves (Fig. 2), and the predictedresponse to thermometry (Fig. 3) and cooling (Fig. 4).
It is instructive to compare the results of this calcu-lation to those of the simple approximation which treats
FIG. 1. Comparison of simulated and measured spin images.Simulated images are on the left. Magnetic field gradients andtemperatures for simulated images are: (a) 0.7 G/cm and 6 nK,(b) 0.06 G/cm and 2 nK, and (c) 0.0024 G/cm and 0.4 nK. Thegradient and fitted temperature for each measured spin image (d)–(f)are similar to the values for the simulated image in the same row. SeeFig. 3 for a comparison of the temperature extracted from this fit tothe real modeled temperature. Note that the total magnetization in allpictures is close to zero.
FIG. 2. Total entropy per particle versus temperature, at variousgradients, for the experimental parameters described in the text.The arrow indicates a possible path followed during adiabatic spingradient demagnetization cooling.
FIG. 3. Ratio between fitted temperature and actual temperatureat two different gradients, assuming perfect imaging. This showsthe effect of corrections due to indistinguishability and unequalscattering lengths. Finite imaging resolution will limit the rangeof temperature that can be measured with any given gradient, asdiscussed in Ref. [13].
FIG. 4. Spin gradient demagnetization cooling. Predicted tem-perature versus final gradient, for several values of the total entropy.
051603-2
RAPID COMMUNICATIONS
THERMOMETRY AND REFRIGERATION IN A TWO- . . . PHYSICAL REVIEW A 82, 051603(R) (2010)
the spin and particle-hole degrees of freedom separately. Inthis approximation, the partition function for an individuallattice site i is assumed to factorize as Z = ZσZ0, whereZσ = ∑
σ exp[−β pσ ·B(xi)], β is 1/kBT , pσ is the magneticmoment of the spin σ , and Z0 is the partition function ofthe particle-hole degrees of freedom (for which see [11,15]).This simple treatment is valid for the case of one atomper lattice site. For occupation number n > 1, there arecorrections which are captured by our more complete model.The first correction arises from a difference U betweenthe mean of the intraspin interaction energies Uσ and theinterspin interaction energy U↑↓. U/Uσ is about 0.007for our states. The leading correction changes the magneticfield gradient at the center of the sample B ′ to an effectivegradient B ′[1 + (n − 1)U/kBT ]. This becomes importantat low temperatures, and destroys the factorizability of thepartition function mentioned above. The second correction isdue to indistinguishability of the atoms. This arises from thequantum mechanical fact that there are three (rather than four)possible spin states for a lattice site with two pseudospin-1/2atoms. The size of both corrections is expected to be small,but in order to treat them fully we have developed the moregeneral model described above.
Under the assumption that Z = ZσZ0, the mean spin 〈s〉 asa function of position, field gradient, and temperature has thesimple form
〈s〉 = tanh[−β p · B(xi)/2], (4)
where p is the difference between the magnetic moments ofthe two states. Spin gradient thermometry is based on the factthat, at finite temperatures, the width of the boundary layer isproportional to the temperature and inversely proportional tothe magnetic field gradient.
The spin profile of Eq. (4) is exact for a 2CMI withone particle per site. The model presented here can be usedto investigate corrections to thermometry at higher filling.Figure 3 shows the temperature measured by fitting to Eq. (4)divided by the actual modeled temperature for two values ofthe gradient. The high-temperature correction is mainly dueto indistinguishability of the atoms, and is only important forsites containing 2 or more atoms. Note that the fitted spinprofile is integrated along both directions transverse to thegradient, so it includes contributions from all occupation-number domains. Although this correction is conceptuallyimportant, it changes the measured temperature by less than15% under our experimental conditions. The correction at lowtemperatures is partly due to the fact that the scattering lengthsU↑↑, U↓↓, and U↑↓ are not all equal. This is expected to result ina curvature of the mixed region between the two spin domains.This curvature arises from a buoyancy effect—the specieswith greater intraspin repulsion will preferentially populate theouter regions of the trap. This effect causes a fit to Eq. (4) tooverestimate the temperature, since a curved boundary appearswider after integration along the directions perpendicular tothe gradient. Another correction at low temperatures arisesif the width of the mixed region becomes much less thanone lattice constant. In this case both the model and the realphysical spin system will not respond measurably to smallchanges in the gradient, and the measured temperature will
overestimate the real temperature. These corrections need tobe taken into account for precision temperature measurementsat extreme temperatures and field gradients, but they do notalter the conclusions of Refs. [13] or [14].
Spin gradient demagnetization cooling is based on thefact that the entropy stored in the mixed region betweenthe two spin domains increases with decreasing magneticfield gradient. If the change of the gradient is adiabatic,then the energy and entropy which flow to the spin degreesof freedom must come from other degrees of freedom, andthe sample’s temperature can be reduced. Although spingradient demagnetization cooling was inspired by (and islocally similar to) adiabatic demagnetization refrigeration incondensed or gaseous systems [18–20], there are importantdifferences between the techniques. Most notably, spin gradi-ent demagnetization cooling varies a magnetic field gradientrather than a homogeneous field, and relies on spin transportrather than spin flips. These differences allow the techniqueto be applied to lattice-trapped ultracold atomic systems. Spingradient demagnetization cooling thus broadens and extendsexisting magnetic refrigeration techniques.
Entropy versus temperature curves such as those plotted inFig. 2 can be used to calculate the response of the system to spingradient demagnetization cooling. If the gradient is reducedperfectly adiabatically, the system will move horizontallyas indicated by the arrow in Fig. 2, and the temperaturewill decrease. This behavior is plotted in Fig. 4 for severalvalues of the total entropy (corresponding to different initialtemperatures). These predictions can be used directly to fitexperimental data, with the initial temperature being the onlyfree parameter. Such fitting gives reasonable agreement (seeRef. [14]).
For sufficiently low initial entropy, the spin degrees offreedom will contain all the entropy in the system whenthe gradient is adiabatically reduced by some factor. Furtherreduction of the gradient below this point is expected tolinearly decrease the temperature of the system until the pointwhere interactions become important. Conversely, if the initialentropy is too high, the spins will become fully disordered atsome finite value of the gradient and will no longer be able toabsorb entropy. Reduction of the gradient below this point willnot change the temperature. This behavior, which is essentiallya finite-size effect, is apparent in the upper curve in Fig. 4. If thegradient is sufficiently high, it can pull the two spin domains sofar apart that the area where they overlap is decreased in size.This effect reduces the entropy capacity of the spin degreesof freedom at high gradients, and is the origin of the slightdownturn in temperatures at the highest gradients in Fig. 4.
Magnetic field gradients of 1 mG/cm are well within therange of the experimentally achievable. Assuming reductionof the gradient from 2 G/cm to 1 mG/cm, our analysis predictsthat samples with an initial entropy lower than about 0.4kB canbe cooled below the spin-ordering temperature. Our modelneglects spin correlations, so the lowest-temperature resultsplotted in Fig. 4 should be taken as evidence that reduction ofthe gradient is capable of cooling below the spin-orderingtemperature rather than a prediction of the dependence oftemperature on gradient below the Curie or Neel temperature.
Figure 5 shows several images of the total entropy dis-tribution at different final gradients during demagnetization.
051603-3
RAPID COMMUNICATIONS
WELD, MIYAKE, MEDLEY, PRITCHARD, AND KETTERLE PHYSICAL REVIEW A 82, 051603(R) (2010)
FIG. 5. (Color online) Entropy distribution during spin gradientdemagnetization cooling. The images are slices of the cloud throughthe center. Each pixel represents one lattice site. All plots are at atotal entropy per particle near 0.3kB , and are thus representative ofthe changing entropy distribution during isentropic demagnetization.Values of the gradient and temperature are as follows: (a) ∇B =0.5 G/cm and T = 3 nK, (b) ∇B = 0.1 G/cm and T = 1.5 nK,(c) ∇B = 0.02 G/cm and T = 0.5 nK. The ring-shaped structures arethe mixed-occupation-number regions between Mott domains whichcarry particle-hole entropy. Note that these regions are narrowerafter reduction of the gradient. This indicates that entropy hasbeen transferred from the mixed-occupation-number regions to themixed-spin region, and the temperature has been reduced.
The pumping of entropy from the kinetic degrees of freedomto the spins is clearly visible.
These theoretical results help elucidate some limitationson and possible extensions to the technique of spin gradientdemagnetization cooling. The technique clearly requires theuse of two states with different magnetic moments—this
excludes, for example, the two lowest states of 7Li at very highfields. The predicted behavior is also, in principle, different forhigher filling factors than it is for n = 1 (although, as discussedabove, we find this effect to be small for the particular caseof 87Rb). For example, strong miscibility or immiscibility ofthe two species would change the response of the system todemagnetization, but only if the maximum occupation numbern is greater than 1 (see also Ref. [21]). The dependence of theresponse to demagnetization on the trap frequencies and totalatom number can also be investigated using the techniquespresented here; the most important effect of varying theseparameters is generally to change the maximum occupationnumber and the spectrum of particle-hole excitations. Forbest cooling performance, the initial entropy should be lowerthan the maximum mixing entropy (kB ln 2 per site for then = 1 case). We believe that spin gradient demagnetizationcooling could, in principle, be applied to fermionic mixturesas well. In fact, the technique does not even require a lattice,and could potentially be applied to the thermal fraction of atrapped two-component gas.
We have presented results of calculations based on a theoret-ical model of the 2CMI and its response to a varying magneticfield gradient. Our results provide quantitative support forspin gradient thermometry and spin gradient demagnetizationcooling.
We thank Eugene Demler, Takuya Kitagawa, David Pekker,and Servaas Kokkelmans for useful and interesting discus-sions. H.M. acknowledges support from the NDSEG fellow-ship program. This work was supported by the NSF, througha MURI program, and under ARO Grant No. W911NF-07-1-0493 with funds from the DARPA OLE program. Corre-spondence and requests for materials should be addressed toD.M.W.
[1] R. Feynman, Int. J. Theor. Phys. 21, 467 (1982).[2] M. Lewenstein et al., Adv. Phys. 56, 243 (2007).[3] I. Bloch, J. Dalibard, and W. Zwerger, Rev. Mod. Phys. 80, 885
(2008).[4] P. A. Lee, N. Nagaosa, and X.-G. Wen, Rev. Mod. Phys. 78, 17
(2006).[5] J. Catani, L. De Sarlo, G. Barontini, F. Minardi, and M. Inguscio,
Phys. Rev. A 77, 011603 (2008).[6] U. Schneider et al., Science 322, 1520 (2008).[7] R. Jordens, N. Strohmaier, K. Gunter, H. Moritz, and
T. Esslinger, Nature (London) 455, 204 (2008).[8] B. Gadway, D. Pertot, R. Reimann, and D. Schneble, Phys. Rev.
Lett. 105, 045303 (2010).[9] S. Ospelkaus et al., Phys. Rev. Lett. 96, 180403 (2006).
[10] L.-M. Duan, E. Demler, and M. D. Lukin, Phys. Rev. Lett. 91,090402 (2003).
[11] T.-L. Ho and Q. Zhou, Phys. Rev. Lett. 99, 120404(2007).
[12] B. Capogrosso-Sansone, S. G. Soyler, N. V. Prokof’ev, andB. V. Svistunov, Phys. Rev. A 81, 053622 (2010).
[13] D. M. Weld et al., Phys. Rev. Lett. 103, 245301(2009).
[14] P. Medley, D. M. Weld, H. Miyake, D. E. Pritchard, andW. Ketterle, e-print arXiv:1006.4674.
[15] F. Gerbier, Phys. Rev. Lett. 99, 120405 (2007).[16] S. S. Natu and E. J. Mueller, e-print arXiv:1005.3090.[17] S. Kokkelmans (private communication).[18] W. F. Giauque, J. Am. Chem. Soc. 49, 1864 (1927).[19] P. Debye, Ann. Phys. 386, 1154 (1926).[20] M. Fattori et al., Nat. Phys. 2, 765 (2006).[21] E. Altman, W. Hofstetter, E. Demler, and M. Lukin, New J.
Phys. 5, 113 (2003).
051603-4
130
Appendix F
Spin Gradient Thermometry for
Ultracold Atoms in Optical
Lattices
This appendix contains a reprint of Ref. [115]: David M. Weld, Patrick Medley,
Hirokazu Miyake, David Hucul, David E. Pritchard, and Wolfgang Ketterle, Spin
Gradient Thermometry for Ultracold Atoms in Optical Lattices, Phys. Rev. Lett.
103, 245301 (2009).
131
Spin Gradient Thermometry for Ultracold Atoms in Optical Lattices
David M. Weld, Patrick Medley, Hirokazu Miyake, David Hucul, David E. Pritchard, and Wolfgang Ketterle
MIT-Harvard Center for Ultracold Atoms, Research Laboratory of Electronics, and Department of Physics,Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA
(Received 20 August 2009; revised manuscript received 10 October 2009; published 7 December 2009)
We demonstrate spin gradient thermometry, a new general method of measuring the temperature of
ultracold atoms in optical lattices. We realize a mixture of spins separated by a magnetic field gradient.
Measurement of the width of the transition layer between the two spin domains serves as a new method of
thermometry which is observed to work over a broad range of lattice depths and temperatures, including in
the Mott insulator regime. We demonstrate the thermometry using ultracold rubidium atoms, and suggest
that interesting spin physics can be realized in this system. The lowest measured temperature is 1 nK,
indicating that the system has reached the quantum regime, where insulating shells are separated by
superfluid layers.
DOI: 10.1103/PhysRevLett.103.245301 PACS numbers: 67.85.d, 03.75.Mn, 05.30.Jp, 75.10.Jm
Ultracold atoms trapped in optical lattices represent anew frontier for the investigation of many-body physics[1,2]. The existence of novel physics at decreasing energyscales drives the quest for lower temperatures in the atomicMott insulator. Insulating Mott shells form at a temperatureT 0:2U, where U is the interaction energy. At the lowertemperature T zJ, where J is the tunneling amplitudeand z is the number of nearest neighbors, the conductinglayers become superfluid and the system enters a quantuminsulator state [3]. At the even colder temperature scaleT J2=U, superexchange-stabilized phases can exist inthe two-component Mott insulator; this is the regime ofquantum magnetism [4]. Various proposals [5,6] have fo-cused on the realization of quantum spin Hamiltonians inthis regime. Detection of superexchange-driven phase tran-sitions in these systems remains a major goal of ultracoldatomic physics. Perhaps the most important barrier toexperimental detection of such a phase transition is therequirement of temperatures well below 1 nK [4].Additional cooling methods [7–10] will be needed to reachthis very interesting temperature scale. However, it is clearthat to assess current methods and to validate future cool-ing techniques, low-temperature thermometry of the Mottinsulator is needed.
Thermometry of systems in the Mott insulating state hasremained a challenge [3,11–14]. In this Letter, we discussand demonstrate a simple and direct method of thermom-etry using a magnetic field gradient which works in thetwo-component Mott insulator.
The theory behind this method of thermometry isstraightforward. The system under consideration is anensemble of atoms in a mixture of two hyperfine statesloaded into a three-dimensional optical lattice in the pres-ence of a weak magnetic field gradient. The two states havedifferent magnetic moments, and are thus pulled towardsopposite sides of the trapped sample by the gradient. At
zero temperature, the spins will segregate completely, anda sharp domain wall will exist between the two spindomains (a small width due to superexchange coupling istypically negligible). This system has the same bulk phys-ics as the single-component Mott insulator, but includesadditional degrees of freedom in the form of spin excita-tions in the domain wall. At finite temperature, spin ex-citations will increase the width of the domain wall. Thiswidth will depend in a simple way on the field gradient, thedifferential Zeeman shift, and the temperature, and canthus be used as a thermometer.For an incoherent mixture of two spins, the partition
function for an individual lattice site can be approximatelyfactorized as Z ¼ ZZ0, where Z ¼ P
expð BðxÞÞ, is 1=kBT, is the magnetic moment of thespin , BðxÞ is the spatially varying magnetic field, andZ0 is the partition function of the particle-hole degrees offreedom (for which see [3]). This approximation is gen-erally valid for the case of one atom per lattice site; foroccupation number n > 1, it is valid when the mean of theintraspin interaction energies U is equal to the interspininteraction energy U"#, which is a good approximation in87Rb [15]. Since the total magnetization is fixed, the aver-age value of the magnetic field is canceled by the corre-sponding Lagrange multiplier; we include this in thedefinition of BðxÞ. We are free to treat the two states ashaving pseudospinþ1 and 1; making that identification,the mean spin hsi as a function of position, gradientstrength, and temperature has the simple form
hsi ¼ tanhð BðxÞ=2Þ; (1)
where is the difference between the magnetic momentsof the two states. A fit of the measured spin distributionwith a function of this form will give the temperature of thesystem. When the Zeeman shift due to the magnetic fieldgradient is a linear function of position, imaging of the spin
PRL 103, 245301 (2009)
Selected for a Viewpoint in PhysicsPHY S I CA L R EV I EW LE T T E R S
week ending11 DECEMBER 2009
0031-9007=09=103(24)=245301(4) 245301-1 2009 The American Physical Society
distribution essentially corresponds to direct imaging ofthe Boltzmann distribution.
The apparatus used to produce ultracold 87Rb atoms isdescribed in Ref. [16]. After cooling, approximately 105
atoms are held in a far-red-detuned crossed optical dipoletrap with trap frequencies between 100 and 200 Hz. Athree-dimensional cubic optical lattice, formed by threeretroreflected beams each of radius 150 m, overlapsthe trapping region. Since spin gradient thermometry doesnot depend on the number of atoms per lattice site n, weperform measurements at a range of n values between 1and 4. The trapping and lattice beams are all derived fromone fiber laser, with a wavelength of 1064 nm. Magneticfield gradients up to a few G=cm can be applied withexternal coils, and calibrated using Stern-Gerlach separa-tion of the different spin states after release from the trap.The gradient is applied along the x direction, which is theweakest axis of the crossed dipole trap. Absorptive imag-ing of the atoms is performed with a camera pointing downalong the vertical z axis.
The sequence of steps used to measure temperatureis as follows. First, a sample of 87Rb atoms in thejF ¼ 1; mF ¼ 1i state is prepared by evaporation in theoptical trap. Here F and mF are the quantum numbers forthe total spin and its projection on the z axis, respectively.The atoms are then placed into a mixture of the j1;1i andj2;2i states by a nonadiabatic magnetic field sweepthrough the microwave transition between the two states.This pair of states was chosen in order to avoid spin-exchange collisions. A magnetic field gradient of2 G=cm is applied along the weak axis of the trap andresults in additional evaporation, which is intended toremove the entropy created by the state preparation [17].At this point, the field gradient is changed to the value to beused for measurement; lower gradients are used for lower-temperature measurements to keep the domain wall widthlarger than the imaging resolution. The optical lattice isthen adiabatically ramped up, typically to a depth of14:5ER, where ER ¼ h2=2m2 is the recoil energy and mis the atomic mass. The transition to the Mott insulatoroccurs at 13:5ER. At this point, the spin structure dependson the temperature as discussed above.
There are several ways to measure the resulting spindistribution. One way is to first take an image of the F ¼ 2atoms in the 14:5ER lattice, then in a second run to illumi-nate the atoms with an optical repumper beam resonantwith the F ¼ 1 to F0 ¼ 2 transition for a few s prior toimaging. This method gives an image of all atoms and animage of just the F ¼ 2 atoms; appropriate subtraction canprovide the spin distribution. It is possible to determine thetemperature from a single image of one spin, but the data inthis Letter were all taken using pairs of images to guardagainst systematic errors.
The temperature can then be measured by fitting the spindistribution to the hyperbolic tangent form. The resulting
thermometer has high dynamic range and variable sensi-tivity, works at all accessible temperatures of interest, andrequires only the simplest fitting procedures.Figure 1 shows data of the type used for spin gradient
thermometry. An image of the total atom density and animage of the spin density are obtained as discussed above.Both images are then integrated along the y direction,which is transverse to the gradient. The spin distribution
is then fit by a function of the form ðxÞ tanhð34BdjBjdx xÞ,
where ðxÞ is the total density distribution. The only freeparameters in this fit are a horizontal and vertical offset andthe temperature T ¼ 1=kB.Figures 2 and 3 show the results of this thermometry on
ultracold 87Rb atoms in an optical lattice. Figure 2 showsthe linear scaling of the inverse width of the domain wall asthe magnetic field gradient is varied while holding thetemperature constant. For widths larger than the opticalresolution, the scaling is as predicted by Eq. (1). The twodata sets plotted in Fig. 2 were taken at two differenttemperatures: 7 and 123 nK, according to the best-fittheoretical lines. Finite optical resolution or motion ofthe atoms during imaging will blur the measured spinprofile and result in an overestimate of the domain wallwidth at high gradients. This effect was modeled by apply-ing a Gaussian blur of radius 4 m to the theoretical 7 nKspin profile at various gradients. The resulting curve, plot-ted as a dash-dotted line in Fig. 2, reproduces the saturationof measured width observed in the experimental data. Theeffect of finite resolution is always to overestimate thetemperature.Figure 3 shows the measured temperature plotted as a
function of the power in the dipole trapping beam whichconfines the atoms in the direction of the magnetic fieldgradient (the x direction). Higher powers in this beam leadto less effective evaporation, and thus higher final tempera-
FIG. 1. Images used for spin gradient thermometry. Data onthe left were taken at a lower optical trap power than data on theright. Panels (a) and (b) are images of the spin distribution.Panels (c) and (d) show the mean spin versus x position. The fitto (c) gives a temperature of 52 nK; the fit to (d) gives atemperature of 296 nK. The inset of (a) shows the axes referredto in the text. The bar in (b) is a size scale.
PRL 103, 245301 (2009) P HY S I CA L R EV I EW LE T T E R Sweek ending
11 DECEMBER 2009
245301-2
tures. As a check of the new method of thermometry, Fig. 3also presents an analysis of the same data using an existingmethod of thermometry, based on measurement of the in-trap width of the atomic cloud along the direction perpen-dicular to the gradient. This second method is based on thewell-known relation 2 ¼ kBT=m!2, where is the 1=e2
half-width of the atomic cloud and ! is the trap frequency
in the direction along which the width is measured [12].The width is determined by a fit to the wings of the trappedcloud. Trap width thermometry is based on a noninteract-ing approximation, and will fail at temperatures less thanUwhen the system starts to become incompressible. As inRef. [12], all points on this plot are in the high-temperaturesingle-band regime (T is less than the band gap but greaterthan the bandwidth). For the temperatures plotted in Fig. 3,the agreement between the two methods is reasonablygood, and gives confidence in the use of spin gradientthermometry in regions of parameter space where no otherthermometer exists.The large dynamic range of spin gradient thermometry
is evident in Fig. 3. Thermometry can be performed attemperatures so high that no condensate exists beforelattice ramp-up. The lowest temperature we have measuredwas achieved by using the new thermometry as a feedbacksignal, enabling adjustment of experimental parameters foroptimization of the final temperature in the Mott insulator.This method allowed us to achieve a measured temperatureas low as 1 nK. At the lattice depth used here, U is 37 nK,and zJ is 6 nK. The measured temperature is thus wellbelow Tc ¼ zJ, the predicted critical temperature for thesuperfluid layer between the n ¼ 1 and n ¼ 0 Mott do-mains. According to the treatment of Ref. [3], at 1 nK thesystem should be well inside the quantum regime, withconcentric quantum insulator shells separated by super-fluid layers. This represents the first direct demonstrationthat this temperature regime has been achieved in the Mottinsulator.At a given value of the magnetic field gradient, very low
temperatures will result in a width of the transition regionsmaller than the imaging optics can resolve (see Fig. 2).However, the width can be increased by decreasing themagnetic field gradient. The lowest measurable tempera-ture will then depend on the minimum achievable gradientas well as the optical resolution, which are technical ratherthan fundamental limitations. In our apparatus, back-ground gradients with all coils turned off are of order103 G=cm, which, given our imaging resolution of afew m, would in principle allow measurement of tem-peratures down to 50 pK or the superexchange scale,whichever is higher.It is instructive to compare the useful range of this new
method of thermometry with that of existing methods. Tofacilitate meaningful comparison with non-lattice-basedmethods, we discuss the range of entropy per particleS=NkB at which a given thermometer works, rather thanthe range of temperature. Condensate fraction thermome-try works for 0:35< S=NkB < 3:5, where the lower limit isset by the difficulty of detecting a thermal fraction less than10%, and the upper limit is set by disappearance of thecondensate. Thermometry based on the thermal cloud sizehas a similar lower bound, but extends to arbitrarily highvalues of S=NkB. Quantitative thermometry based on the
FIG. 3. Validation of spin gradient thermometry. Comparisonof two measured temperatures versus final power in one of theoptical trapping beams. Squares represent the results of in-trapcloud width thermometry, and circles represent the results ofspin gradient thermometry (see text for details). Error barsrepresent estimated uncertainties. The dashed line is a linear fitto the spin gradient thermometry data. The closeness of this fitsuggests that the temperature reached is proportional to the trapdepth.
FIG. 2. Independence of the measured temperature on theapplied field gradient. The inverse of the width of the spin profileis plotted as a function of magnetic field gradient for two datasets at two different temperatures. For constant temperature, alinear curve is expected. The width is defined as the distancefrom the center to the position where the mean spin is 1=2. Thesolid (dashed) line assumes a temperature of 123 nK (7 nK) andperfect imaging. The measured width of the colder data setsaturates at high gradient because of finite imaging resolution.The dotted line assumes a temperature of 7 nK and an imagingresolution of 4 m.
PRL 103, 245301 (2009) P HY S I CA L R EV I EW LE T T E R Sweek ending
11 DECEMBER 2009
245301-3
visibility of interference peaks upon release from the lat-tice requires state-of-the-art quantum Monte Carlo calcu-lations fitted to the data. This technique was recently usedto measure temperatures as low as 0:08U in the superfluidphase near the Mott insulator transition [18]. This methodcannot be applied deep in the Mott insulating state [11].Measurement of the width of the conducting layers be-tween the Mott shells is the only previously proposedmethod which works directly in the Mott insulating state[3,4,19]. However, this method requires tomographic tech-niques, and the useful range of entropy is rather narrow:0:4< S=NkB < lnð2Þ, where the upper limit is set by themelting of the Mott shells, and the lower limit is anestimate based on typical trapping parameters and opticalresolution. Counting only spin excitations, the range ofspin entropy per particle at which spin gradient thermom-etry works in our system is 0:1< S=NkB < lnð2Þ, wherethe lower limit is a function of optical resolution andsample size and the upper limit corresponds to the pointat which the domain wall becomes as wide as the sample. Itis important to note that spin gradient thermometry canwork even if the entropy of the particle-hole excitationslies outside of this range in either direction. For example,spin gradient thermometry can work at arbitrarily highvalues of the total entropy per particle S=NkB, assumingthe field gradient is increased to the point where S=NkB <lnð2Þ.
The method of thermometry presented here works be-cause the two-component Mott insulator in a field gradienthas a spectrum of soft and easily measurable spin excita-tions. The wide dynamic range of this method is a result ofthe fact that, in contrast to the gapped spectrum of the bulkone-component Mott insulator, the energy of the spin ex-citations can be tuned by adjusting the strength of themagnetic field gradient. The addition of a field gradientand a second spin component does not change the bulkproperties of the Mott insulator and can be regarded as‘‘attaching’’ a general thermometer to the first component.
The two-component Mott insulator in a field gradient isa rich system which can provide experimental access tonovel spin physics as well as thermometry. In the workpresented here, we have always allowed the spin distribu-tion to equilibrate in the gradient before ramping up theoptical lattice. However, changing the gradient after theatoms were already loaded into the lattice should open upseveral interesting scientific opportunities, in which thegradient is used to manipulate or perturb the atoms ratherthan as a diagnostic tool. If, for example, the gradient weresuddenly changed after lattice ramp-up, one could probenonequilibrium spin dynamics in a many-body quantumsystem. If the gradient were instead lowered adiabaticallyafter ramp-up, adiabatic cooling of theMott insulator could
potentially be performed which, in contrast to [20], wouldnot involve spin-flip collisions.In conclusion, we have proposed and demonstrated a
new method of thermometry for ultracold atoms in opticallattices. We have used the new method to measure tem-peratures in the Mott insulator as low as 1 nK. Thistemperature is to the best of our knowledge the lowestever measured in a lattice, and it indicates that the systemis deep in the quantum Mott regime.The authors thank Eugene Demler, Takuya Kitagawa,
David Pekker, and Lode Pollet for helpful discussions. Thiswork was supported by the NSF, through a MURI program,and under ARO Grant No. W911NF-07-1-0493 with fundsfrom the DARPA OLE program. H.M. acknowledges sup-port from the NSF.
[1] I. Bloch, J. Dalibard, and W. Zwerger, Rev. Mod. Phys. 80,885 (2008).
[2] M. Lewenstein et al., Adv. Phys. 56, 243 (2007).[3] F. Gerbier, Phys. Rev. Lett. 99, 120405 (2007).[4] T.-L. Ho and Q. Zhou, Phys. Rev. Lett. 99, 120404 (2007).[5] L. Duan, E. Demler, and M. Lukin, Phys. Rev. Lett. 91,
090402 (2003).[6] E. Altman, W. Hofstetter, E. Demler, and M. Lukin,
New J. Phys. 5, 113 (2003).[7] M. Popp, J.-J. Garcia-Ripoll, K.G. Vollbrecht, and J. I.
Cirac, Phys. Rev. A 74, 013622 (2006).[8] P. Rabl, A. J. Daley, P. O. Fedichev, J. I. Cirac, and
P. Zoller, Phys. Rev. Lett. 91, 110403 (2003).[9] J.-S. Bernier et al., Phys. Rev. A 79, 061601 (2009).[10] T.-L. Ho and Q. Zhou, Proc. Natl. Acad. Sci. U.S.A. 106,
6916 (2009).[11] L. Pollet, C. Kollath, K.V. Houcke, and M. Troyer, New J.
Phys. 10, 065001 (2008).[12] D. McKay, M. White, and B. DeMarco, Phys. Rev. A 79,
063605 (2009).[13] A. Hoffmann and A. Pelster, Phys. Rev. A 79, 053623
(2009).[14] B. Capogrosso-Sansone, E. Kozik, N. Prokof’ev, and
B. Svistunov, Phys. Rev. A 75, 013619 (2007).[15] For other atomic species, this treatment can readily be
extended to include effects of finite U ¼ U U"#. Theleading correction changes the magnetic field gradient atthe center of the sample B0 to an effective gradient B0ð1þðn 1ÞU=kBTÞ.
[16] E.W. Streed et al., Rev. Sci. Instrum. 77, 023106 (2006).[17] C.-L. Hung, X. Zhang, N. Gemelke, and C. Chin, Phys.
Rev. A 78, 011604 (2008).[18] S. Trotzky et al., arXiv:0905.4882.[19] S. Folling, A. Widera, T. Muller, F. Gerbier, and I. Bloch,
Phys. Rev. Lett. 97, 060403 (2006).[20] M. Fattori et al., Nature Phys. 2, 765 (2006).
PRL 103, 245301 (2009) P HY S I CA L R EV I EW LE T T E R Sweek ending
11 DECEMBER 2009
245301-4
136
Bibliography
[1] J. R. Abo-Shaeer, C. Raman, J. M. Vogels, and W. Ketterle. Observation ofvortex lattices in bose-einstein condensates. Science, 292(5516):476–479, 2001.
[2] Y. Aharonov and D. Bohm. Significance of electromagnetic potentials in thequantum theory. Phys. Rev., 115:485–491, Aug 1959.
[3] M. Aidelsburger, M. Atala, S. Nascimbene, S. Trotzky, Y.-A. Chen, and I. Bloch.Experimental realization of strong effective magnetic fields in an optical lattice.Phys. Rev. Lett., 107:255301, Dec 2011.
[4] K. Aikawa, A. Frisch, M. Mark, S. Baier, A. Rietzler, R. Grimm, and F. Ferlaino.Bose-einstein condensation of erbium. Phys. Rev. Lett., 108:210401, May 2012.
[5] A. Alberti, G. Ferrari, V. V. Ivanov, M. L. Chiofalo, and G. M. Tino. Atomicwave packets in amplitude-modulated vertical optical lattices. New J. Phys.,12(6):065037, 2010.
[6] A. Alberti, V. V. Ivanov, G. M. Tino, and G. Ferrari. Engineering the quantumtransport of atomic wavefunctions over macroscopic distances. Nat Phys, 5:547,August 2009.
[7] M. H. Anderson, J. R. Ensher, M. R. Matthews, C. E. Wieman, and E. A.Cornell. Observation of bose-einstein condensation in a dilute atomic vapor.Science, 269(5221):198–201, 1995.
[8] M. R. Andrews, C. G. Townsend, H.-J. Miesner, D. S. Durfee, D. M. Kurn,and W. Ketterle. Observation of interference between two bose condensates.Science, 275(5300):637–641, 1997.
[9] N. Ashcroft and N. D. Mermin. Solid State Physics. Brooks Cole, Belmont,Massachusetts, 1976.
[10] A. Auerbach. Interacting Electrons and Quantum Magnetism. Springer-Verlag,New York City, New York, 1994.
[11] Waseem S. Bakr, Jonathon I. Gillen, Amy Peng, Simon Folling, and MarkusGreiner. A quantum gas microscope for detecting single atoms in a hubbard-regime optical lattice. Nature, 462:74, November 2009.
137
[12] G. Birkl, M. Gatzke, I. H. Deutsch, S. L. Rolston, and W. D. Phillips. Braggscattering from atoms in optical lattices. Phys. Rev. Lett., 75:2823–2826, Oct1995.
[13] Immanuel Bloch, Jean Dalibard, and Wilhelm Zwerger. Many-body physicswith ultracold gases. Rev. Mod. Phys., 80:885–964, Jul 2008.
[14] B. N. Brockhouse. Scattering of neutrons by spin waves in magnetite. Phys.Rev., 106:859–864, Jun 1957.
[15] B. Capogrosso-Sansone, S. G. Soyler, N. V. Prokof’ev, and B. V. Svistunov.Critical entropies for magnetic ordering in bosonic mixtures on a lattice. Phys.Rev. A, 81:053622, May 2010.
[16] Michael S. Chapman, Christopher R. Ekstrom, Troy D. Hammond, JorgSchmiedmayer, Bridget E. Tannian, Stefan Wehinger, and David E. Pritchard.Near-field imaging of atom diffraction gratings: The atomic talbot effect. Phys.Rev. A, 51:R14–R17, Jan 1995.
[17] Lawrence W. Cheuk, Ariel T. Sommer, Zoran Hadzibabic, Tarik Yefsah,Waseem S. Bakr, and Martin W. Zwierlein. Spin-injection spectroscopy of aspin-orbit coupled fermi gas. Phys. Rev. Lett., 109:095302, Aug 2012.
[18] Adrian Cho. The mad dash to make light crystals. Science, 320(5874):312–313,2008.
[19] Steven Chu, J. E. Bjorkholm, A. Ashkin, and A. Cable. Experimental observa-tion of optically trapped atoms. Phys. Rev. Lett., 57:314–317, Jul 1986.
[20] Leon N. Cooper. Bound electron pairs in a degenerate fermi gas. Phys. Rev.,104:1189–1190, Nov 1956.
[21] T. A. Corcovilos, S. K. Baur, J. M. Hitchcock, E. J. Mueller, and R. G. Hulet.Detecting antiferromagnetism of atoms in an optical lattice via optical braggscattering. Phys. Rev. A, 81:013415, Jan 2010.
[22] Jean Dalibard, Fabrice Gerbier, Gediminas Juzeliunas, and Patrik Ohberg.Colloquium: Artificial gauge potentials for neutral atoms. Rev. Mod. Phys.,83:1523–1543, Nov 2011.
[23] K. B. Davis, M. O. Mewes, M. R. Andrews, N. J. van Druten, D. S. Durfee,D. M. Kurn, and W. Ketterle. Bose-einstein condensation in a gas of sodiumatoms. Phys. Rev. Lett., 75:3969–3973, Nov 1995.
[24] J. H. de Boer and E. J. W. Verwey. Semi-conductors with partially andwith completely filled 3 d -lattice bands. Proceedings of the Physical Society,49(4S):59, 1937.
138
[25] Y. N. Martinez de Escobar, P. G. Mickelson, M. Yan, B. J. DeSalvo, S. B.Nagel, and T. C. Killian. Bose-einstein condensation of 84Sr. Phys. Rev. Lett.,103:200402, Nov 2009.
[26] C. R. Dean, L. Wang, P. Maher, C. Forsythe, F. Ghahari, Y. Gao, J. Katoch,M. Ishigami, P. Moon, M. Koshino, T. Taniguchi, K. Watanabe, K. L. Shepard,J. Hone, and P. Kim. Hofstadter’s butterfly in moire superlattices: A fractalquantum Hall effect, arXiv:1212.4783 [cond-mat.mes-hall], December 2012.
[27] B. DeMarco and D. S. Jin. Onset of fermi degeneracy in a trapped atomic gas.Science, 285(5434):1703–1706, 1999.
[28] L. Deng, E. W. Hagley, J. Denschlag, J. E. Simsarian, Mark Edwards,Charles W. Clark, K. Helmerson, S. L. Rolston, and W. D. Phillips. Temporal,matter-wave-dispersion talbot effect. Phys. Rev. Lett., 83:5407–5411, Dec 1999.
[29] L.-M. Duan, E. Demler, and M. D. Lukin. Controlling spin exchange interac-tions of ultracold atoms in optical lattices. Phys. Rev. Lett., 91:090402, Aug2003.
[30] W. Ehrenberg and R. E. Siday. The refractive index in electron optics and theprinciples of dynamics. Proc. Phys. Soc. Lond., 62(1):8, 1949.
[31] Manuel Endres, Takeshi Fukuhara, David Pekker, Marc Cheneau, Peter Schauß,Christian Gross, Eugene Demler, Stefan Kuhr, and Immanuel Bloch. The ‘higgs’amplitude mode at the two-dimensional superfluid/mott insulator transition.Nature, 487:454, July 2012.
[32] R. P. Feynman. Simulating physics with computers. International Journal ofTheoretical Physics, 21:467, 1982.
[33] Matthew P. A. Fisher, Peter B. Weichman, G. Grinstein, and Daniel S. Fisher.Boson localization and the superfluid-insulator transition. Phys. Rev. B, 40:546–570, Jul 1989.
[34] Alan Gallagher and David E. Pritchard. Exoergic collisions of cold na∗-na.Phys. Rev. Lett., 63:957–960, Aug 1989.
[35] Thomas R. Gentile, Barbara J. Hughey, Daniel Kleppner, and Theodore W.Ducas. Experimental study of one- and two-photon rabi oscillations. Phys.Rev. A, 40:5103–5115, Nov 1989.
[36] Fabrice Gerbier. Boson mott insulators at finite temperatures. Phys. Rev. Lett.,99:120405, Sep 2007.
[37] Fabrice Gerbier and Jean Dalibard. Gauge fields for ultracold atoms in opticalsuperlattices. New Journal of Physics, 12:033007, March 2010.
139
[38] Fabrice Gerbier, Artur Widera, Simon Folling, Olaf Mandel, Tatjana Gericke,and Immanuel Bloch. Phase coherence of an atomic mott insulator. Phys. Rev.Lett., 95:050404, Jul 2005.
[39] Thierry Giamarchi, Christian Ruegg, and Oleg Tchernyshyov. Bose-einsteincondensation in magnetic insulators. Nat Phys, 4:198, March 2008.
[40] Markus Gluck, Andrey R. Kolovsky, and Hans Jurgen Korsch. Wannier-stark resonances in optical and semiconductor superlattices. Physics Reports,366(3):103 – 182, 2002.
[41] A. Gorlitz, J. M. Vogels, A. E. Leanhardt, C. Raman, T. L. Gustavson, J. R.Abo-Shaeer, A. P. Chikkatur, S. Gupta, S. Inouye, T. Rosenband, and W. Ket-terle. Realization of bose-einstein condensates in lower dimensions. Phys. Rev.Lett., 87:130402, Sep 2001.
[42] Markus Greiner, Olaf Mandel, Tilman Esslinger, Theodor W. Hansch, and Im-manuel Bloch. Quantum phase transition from a superfluid to a mott insulatorin a gas of ultracold atoms. Nature, 415:39, January 2002.
[43] Axel Griesmaier, Jorg Werner, Sven Hensler, Jurgen Stuhler, and Tilman Pfau.Bose-einstein condensation of chromium. Phys. Rev. Lett., 94:160401, Apr 2005.
[44] D. J. Griffiths. Introduction to Quantum Mechanics. Pearson Prentice Hall,Upper Saddle River, NJ, 2005.
[45] Milena Grifoni and Peter Hanggi. Driven quantum tunneling. Physics Reports,304(5 - 6):229 – 354, 1998.
[46] Subhadeep Gupta, Aaron E. Leanhardt, Alexander D. Cronin, and David E.Pritchard. Coherent manipulation of atoms with standing light waves. ComptesRendus de l’Academie des Sciences - Series IV - Physics, 2(3):479 – 495, 2001.
[47] Zoran Hadzibabic, Peter Kruger, Marc Cheneau, Baptiste Battelier, and JeanDalibard. Berezinskii-kosterlitz-thouless crossover in a trapped atomic gas. Na-ture, 441:1118, June 2006.
[48] M. Hafezi, A. S. Sørensen, E. Demler, and M. D. Lukin. Fractional quantumhall effect in optical lattices. Phys. Rev. A, 76:023613, Aug 2007.
[49] P. G. Harper. Single band motion of conduction electrons in a uniform magneticfield. Proc. Phys. Soc. A, 68:874–878, Oct 1955.
[50] A. C. Hewson. The Kondo Problem to Heavy Fermions. Cambridge UniversityPress, Cambridge, England, 1997.
[51] Tin-Lun Ho and Qi Zhou. Intrinsic heating and cooling in adiabatic processesfor bosons in optical lattices. Phys. Rev. Lett., 99:120404, Sep 2007.
140
[52] Douglas R. Hofstadter. Energy levels and wave functions of bloch electrons inrational and irrational magnetic fields. Phys. Rev. B, 14:2239–2249, Sep 1976.
[53] J. Hubbard. Electron correlations in narrow energy bands. Proceedings ofthe Royal Society of London. Series A. Mathematical and Physical Sciences,276(1365):238–257, 1963.
[54] B. Hunt, J. D. Sanchez-Yamagishi, A. F. Young, K. Watanabe, T. Taniguchi,P. Moon, M. Koshino, P. Jarillo-Herrero, and R. C. Ashoori. MassiveDirac fermions and Hofstadter butterfly in a van der Waals heterostructure,arXiv:1303.6942 [cond-mat.mes-hall], March 2013.
[55] S. Inouye, M. R. Andrews, J. Stenger, H.-J. Miesner, D. M. Stamper-Kurn, andW. Ketterle. Observation of feshbach resonances in a bose-einstein condensate.Nature, 392:151, March 1998.
[56] V. V. Ivanov, A. Alberti, M. Schioppo, G. Ferrari, M. Artoni, M. L. Chiofalo,and G. M. Tino. Coherent delocalization of atomic wave packets in drivenlattice potentials. Phys. Rev. Lett., 100:043602, Jan 2008.
[57] J. D. Jackson. From lorenz to coulomb and other explicit gauge transformations.American Journal of Physics, 70(9):917–928, 2002.
[58] D. Jaksch, C. Bruder, J. I. Cirac, C. W. Gardiner, and P. Zoller. Cold bosonicatoms in optical lattices. Phys. Rev. Lett., 81:3108–3111, Oct 1998.
[59] D. Jaksch and P. Zoller. Creation of effective magnetic fields in optical lattices:the hofstadter butterfly for cold neutral atoms. New J. Phys., 5:56, May 2003.
[60] K. Jimenez-Garcıa, L. J. LeBlanc, R. A. Williams, M. C. Beeler, A. R. Perry,and I. B. Spielman. Peierls substitution in an engineered lattice potential. Phys.Rev. Lett., 108:225303, May 2012.
[61] Robert Jordens, Niels Strohmaier, Kenneth Gunter, Henning Moritz, andTilman Esslinger. A mott insulator of fermionic atoms in an optical lattice.Nature, 455:204, September 2008.
[62] W Ketterle. Bose-einstein condensation in dilute atomic gases: atomic physicsmeets condensed matter physics. Physica B: Condensed Matter, 280(1-4):11 –19, 2000.
[63] D. I. Khomskii. Basic Aspects of the Quantum Theory of Solids. CambridgeUniversity Press, Cambridge, England, 2010.
[64] A.Yu. Kitaev. Fault-tolerant quantum computation by anyons. Annals ofPhysics, 303(1):2 – 30, 2003.
[65] W. Kohn. Analytic properties of bloch waves and wannier functions. Phys.Rev., 115:809–821, Aug 1959.
141
[66] Sebastian Kraft, Felix Vogt, Oliver Appel, Fritz Riehle, and Uwe Sterr.Bose-einstein condensation of alkaline earth atoms: 40Ca. Phys. Rev. Lett.,103:130401, Sep 2009.
[67] A. B. Kuklov and B. V. Svistunov. Counterflow superfluidity of two-speciesultracold atoms in a commensurate optical lattice. Phys. Rev. Lett., 90:100401,Mar 2003.
[68] R. B. Laughlin. Anomalous quantum hall effect: An incompressible quantumfluid with fractionally charged excitations. Phys. Rev. Lett., 50:1395–1398, May1983.
[69] Patrick A. Lee, Naoto Nagaosa, and Xiao-Gang Wen. Doping a mott insulator:Physics of high-temperature superconductivity. Rev. Mod. Phys., 78:17–85, Jan2006.
[70] Maciej Lewenstein, Anna Sanpera, Veronica Ahufinger, Bogdan Damski, AditiSen(De), and Ujjwal Sen. Ultracold atomic gases in optical lattices: mimickingcondensed matter physics and beyond. Advances in Physics, 56(2):243–379,2007.
[71] H. Lignier, C. Sias, D. Ciampini, Y. Singh, A. Zenesini, O. Morsch, and E. Ari-mondo. Dynamical control of matter-wave tunneling in periodic potentials.Phys. Rev. Lett., 99:220403, Nov 2007.
[72] Y.-J. Lin, R. L. Compton, K. Jimenez-Garcia, J. V. Porto, and I. B. Spiel-man. Synthetic magnetic fields for ultracold neutral atoms. Nature, 462:628,December 2009.
[73] Mingwu Lu, Nathaniel Q. Burdick, Seo Ho Youn, and Benjamin L. Lev. Stronglydipolar bose-einstein condensate of dysprosium. Phys. Rev. Lett., 107:190401,Oct 2011.
[74] K. W. Madison, F. Chevy, W. Wohlleben, and J. Dalibard. Vortex formationin a stirred bose-einstein condensate. Phys. Rev. Lett., 84:806–809, Jan 2000.
[75] K. W. Madison, M. C. Fischer, R. B. Diener, Qian Niu, and M. G. Raizen. Dy-namical bloch band suppression in an optical lattice. Phys. Rev. Lett., 81:5093–5096, Dec 1998.
[76] Patrick Medley, David M. Weld, Hirokazu Miyake, David E. Pritchard, andWolfgang Ketterle. Spin gradient demagnetization cooling of ultracold atoms.Phys. Rev. Lett., 106:195301, May 2011.
[77] Hirokazu Miyake, Georgios A. Siviloglou, Graciana Puentes, David E.Pritchard, Wolfgang Ketterle, and David M. Weld. Bragg scattering as a probeof atomic wave functions and quantum phase transitions in optical lattices.Phys. Rev. Lett., 107:175302, Oct 2011.
142
[78] G. Moller and N. R. Cooper. Condensed ground states of frustrated bose-hubbard models. Phys. Rev. A, 82:063625, Dec 2010.
[79] N. F. Mott. The basis of the electron theory of metals, with special reference tothe transition metals. Proceedings of the Physical Society. Section A, 62(7):416,1949.
[80] Yu. B. Ovchinnikov, J. H. Muller, M. R. Doery, E. J. D. Vredenbregt, K. Helmer-son, S. L. Rolston, and W. D. Phillips. Diffraction of a released bose-einsteincondensate by a pulsed standing light wave. Phys. Rev. Lett., 83:284–287, Jul1999.
[81] M. E. Peskin and D. V. Schroeder. An Introduction to Quantum Field Theory.Westview Press, 1995.
[82] T. P. Polak and T. A. Zaleski. Time-of-flight patterns of ultracold bosons inoptical lattices in various abelian artificial magnetic field gauges. Phys. Rev. A,87:033614, Mar 2013.
[83] Stephen Powell, Ryan Barnett, Rajdeep Sensarma, and Sankar Das Sarma.Bogoliubov theory of interacting bosons on a lattice in a synthetic magneticfield. Phys. Rev. A, 83:013612, Jan 2011.
[84] G. Raithel, G. Birkl, A. Kastberg, W. D. Phillips, and S. L. Rolston. Coolingand localization dynamics in optical lattices. Phys. Rev. Lett., 78:630–633, Jan1997.
[85] Subir Sachdev. Quantum magnetism and criticality. Nat Phys, 4:173, March2008.
[86] J. J. Sakurai. Modern Quantum Mechanics. Addison-Wesley Publishing Com-pany, Reading, MA, 1994.
[87] L. I. Schiff. Quantum Mechanics. McGraw-Hill, New York, 1968.
[88] U. Schneider, L. Hackermller, S. Will, Th. Best, I. Bloch, T. A. Costi, R. W.Helmes, D. Rasch, and A. Rosch. Metallic and insulating phases of repulsivelyinteracting fermions in a 3d optical lattice. Science, 322(5907):1520–1525, 2008.
[89] J. R. Schrieffer and P. A. Wolff. Relation between the anderson and kondohamiltonians. Phys. Rev., 149:491–492, Sep 1966.
[90] Jacob F. Sherson, Christof Weitenberg, Manuel Endres, Marc Cheneau, Im-manuel Bloch, and Stefan Kuhr. Single-atom-resolved fluorescence imaging ofan atomic mott insulator. Nature, 467:68, September 2010.
[91] C. G. Shull and J. Samuel Smart. Detection of antiferromagnetism by neutrondiffraction. Phys. Rev., 76:1256–1257, Oct 1949.
143
[92] C. Sias, H. Lignier, Y. P. Singh, A. Zenesini, D. Ciampini, O. Morsch, andE. Arimondo. Observation of photon-assisted tunneling in optical lattices. Phys.Rev. Lett., 100:040404, Feb 2008.
[93] Jonathan Simon, Waseem S. Bakr, Ruichao Ma, M. Eric Tai, Philipp M. Preiss,and Markus Greiner. Quantum simulation of antiferromagnetic spin chains inan optical lattice. Nature, 472:307, April 2011.
[94] Anders S. Sørensen, Eugene Demler, and Mikhail D. Lukin. Fractional quantumhall states of atoms in optical lattices. Phys. Rev. Lett., 94:086803, Mar 2005.
[95] D. M. Stamper-Kurn, M. R. Andrews, A. P. Chikkatur, S. Inouye, H.-J. Miesner,J. Stenger, and W. Ketterle. Optical confinement of a bose-einstein condensate.Phys. Rev. Lett., 80:2027–2030, Mar 1998.
[96] D. M. Stamper-Kurn, M. R. Andrews, A. P. Chikkatur, S. Inouye, H.-J. Miesner,J. Stenger, and W. Ketterle. Optical confinement of a bose-einstein condensate.Phys. Rev. Lett., 80:2027–2030, Mar 1998.
[97] Daniel A. Steck. “Rubidium 87 D Line Data”, available online athttp://steck.us/alkalidata (revision 2.1.4, 23 December 2010).
[98] Simon Stellmer, Meng Khoon Tey, Bo Huang, Rudolf Grimm, and FlorianSchreck. Bose-einstein condensation of strontium. Phys. Rev. Lett., 103:200401,Nov 2009.
[99] J. Stenger, S. Inouye, D. M. Stamper-Kurn, H.-J. Miesner, A. P. Chikkatur, andW. Ketterle. Spin domains in ground-state bose-einstein condensates. Nature,396:345, November 1998.
[100] Erik W. Streed, Ananth P. Chikkatur, Todd L. Gustavson, Micah Boyd, YoshioTorii, Dominik Schneble, Gretchen K. Campbell, David E. Pritchard, and Wolf-gang Ketterle. Large atom number bose-einstein condensate machines. Reviewof Scientific Instruments, 77(2):023106, 2006.
[101] J. Struck, C. Olschlager, M. Weinberg, P. Hauke, J. Simonet, A. Eckardt,M. Lewenstein, K. Sengstock, and P. Windpassinger. Tunable gauge poten-tial for neutral and spinless particles in driven optical lattices. Phys. Rev. Lett.,108:225304, May 2012.
[102] Yosuke Takasu, Kenichi Maki, Kaduki Komori, Tetsushi Takano, KazuhitoHonda, Mitsutaka Kumakura, Tsutomu Yabuzaki, and Yoshiro Takahashi.Spin-singlet bose-einstein condensation of two-electron atoms. Phys. Rev. Lett.,91:040404, Jul 2003.
[103] M. G. Tarallo, A. Alberti, N. Poli, M. L. Chiofalo, F.-Y. Wang, and G. M.Tino. Delocalization-enhanced bloch oscillations and driven resonant tunnelingin optical lattices for precision force measurements. Phys. Rev. A, 86:033615,Sep 2012.
144
[104] Leticia Tarruell, Daniel Greif, Thomas Uehlinger, Gregor Jotzu, and TilmanEsslinger. Creating, moving and merging dirac points with a fermi gas in atunable honeycomb lattice. Nature, 483:302, March 2012.
[105] D. J. Thouless, M. Kohmoto, M. P. Nightingale, and M. den Nijs. Quantizedhall conductance in a two-dimensional periodic potential. Phys. Rev. Lett.,49:405–408, Aug 1982.
[106] S. Trotzky, L. Pollet, F. Gerbier, U. Schnorrberger, I. Bloch, N. V. Prokof/’ev,B. Svistunov, and M. Troyer. Suppression of the critical temperature for super-fluidity near the mott transition. Nat Phys, 6:998, December 2010.
[107] D. C. Tsui, H. L. Stormer, and A. C. Gossard. Two-dimensional magneto-transport in the extreme quantum limit. Phys. Rev. Lett., 48:1559–1562, May1982.
[108] D. van Oosten, P. van der Straten, and H. T. C. Stoof. Quantum phases in anoptical lattice. Phys. Rev. A, 63:053601, Apr 2001.
[109] A. Wachter and H. Hoeber. Compendium of Theoretical Physics. Springer, NewYork, 2006.
[110] Pengjun Wang, Zeng-Qiang Yu, Zhengkun Fu, Jiao Miao, Lianghui Huang,Shijie Chai, Hui Zhai, and Jing Zhang. Spin-orbit coupled degenerate fermigases. Phys. Rev. Lett., 109:095301, Aug 2012.
[111] J. D. Watson and F. H. C. Crick. Molecular structure of nucleic acids: Astructure for deoxyribose nucleic acid. Nature, 171:737, April 1953.
[112] Matthias Weidemuller, Axel Gorlitz, Theodor W. Hansch, and Andreas Hem-merich. Local and global properties of light-bound atomic lattices investigatedby bragg diffraction. Phys. Rev. A, 58:4647–4661, Dec 1998.
[113] Matthias Weidemuller, Andreas Hemmerich, Axel Gorlitz, Tilman Esslinger,and Theodor W. Hansch. Bragg diffraction in an atomic lattice bound by light.Phys. Rev. Lett., 75:4583–4586, Dec 1995.
[114] Christof Weitenberg, Peter Schauß, Takeshi Fukuhara, Marc Cheneau, ManuelEndres, Immanuel Bloch, and Stefan Kuhr. Coherent light scattering from atwo-dimensional mott insulator. Phys. Rev. Lett., 106:215301, May 2011.
[115] David M. Weld, Patrick Medley, Hirokazu Miyake, David Hucul, David E.Pritchard, and Wolfgang Ketterle. Spin gradient thermometry for ultracoldatoms in optical lattices. Phys. Rev. Lett., 103:245301, Dec 2009.
[116] David M. Weld, Hirokazu Miyake, Patrick Medley, David E. Pritchard, andWolfgang Ketterle. Thermometry and refrigeration in a two-component mottinsulator of ultracold atoms. Phys. Rev. A, 82:051603, Nov 2010.
145
[117] K. C. Wright, R. B. Blakestad, C. J. Lobb, W. D. Phillips, and G. K. Campbell.Driving phase slips in a superfluid atom circuit with a rotating weak link. Phys.Rev. Lett., 110:025302, Jan 2013.
[118] K. Xu, Y. Liu, D. E. Miller, J. K. Chin, W. Setiawan, and W. Ketterle. Ob-servation of strong quantum depletion in a gaseous bose-einstein condensate.Phys. Rev. Lett., 96:180405, May 2006.
[119] Sen Yang, A. T. Hammack, M. M. Fogler, L. V. Butov, and A. C. Gossard.Coherence length of cold exciton gases in coupled quantum wells. Phys. Rev.Lett., 97:187402, Oct 2006.
[120] D. R. Yennie. Integral quantum hall effect for nonspecialists. Rev. Mod. Phys.,59:781–824, Jul 1987.
[121] J. Zak. Magnetic translation group. Phys. Rev., 134:A1602–A1606, Jun 1964.
[122] J. Zak. Magnetic translation group. ii. irreducible representations. Phys. Rev.,134:A1607–A1611, Jun 1964.
[123] M. W. Zwierlein, J. R. Abo-Shaeer, A. Schirotzek, C. H. Schunck, and W. Ket-terle. Vortices and superfluidity in a strongly interacting fermi gas. Nature,435:1047, June 2005.
146